YT YouTube2Text

YouTube2Text → Transcripts → Day 2, Statistics, Ch1 sec2 to Ch1 sec5

Day 2, Statistics, Ch1 sec2 to Ch1 sec5 — Transcript

by Eduardo Barajas · 16,991 words · 3,062 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00okay so more fun with chapter one
0:03on uh defining some of these things
0:07uh we left off with um
0:10uh cross-sectional studies case control
0:13studies
0:13cohort studies right so we're looking at
0:16um
0:17observational studies versus designed
0:19experiments
0:20um a designed experiment
0:23um is when you
0:26think about in a designed experiment
0:29it's important to know that you're
0:30intentionally
0:32involving yourself you're you're
0:34manipulating something
0:36uh and determining how that manipulation
0:39is affecting your results uh so that's
0:42really the
0:42the key difference between a designed
0:45experiment and observational studies in
0:47observational study you just
0:49sit back and watch what happens right
0:52so um if you wanted to
0:56think about sodas for example um
0:59in my opinion i think that in 30 years
1:02we'll look at sodas kind of the same way
1:06we looked at tobacco in the 60s and 70s
1:10people just were convinced that there
1:11was nothing wrong with them and they
1:12were fine and
1:13in some ways they were kind of helpful
1:15and they kind of give you a pep in your
1:17step and
1:18that kind of thing then but then later
1:20it turned out that they're actually
1:21really terrible
1:22and they you know give you a lot of
1:23medical problems
1:25personally i think that's what's gonna
1:26happen with soda uh that right now
1:29we don't really think of it as that big
1:31of a thing
1:32um but i think in 30 years they'll
1:36probably figure out the sodas are
1:37probably really bad for you
1:40and lead to diabetes and all kinds of
1:42problems
1:43anyway how do we comp how do we convince
1:45people of that how do we
1:47figure out if that's true right so in an
1:50observational study we would just watch
1:53right we just look around let's pick a
1:55thousand people
1:57and observe them let's see how often
2:00they consume
2:01soda let's see their habits let's keep
2:03track of their health records
2:05um and then wait 20 years
2:09and see if those people that drank a lot
2:12of soda
2:13end up having uh more medical problems
2:16right that would be an observational
2:18study
2:19um on the other hand if it was an
2:21experiment what we would do
2:23is we would take a random group of
2:25people we would randomly assign them to
2:28two different groups
2:29one group we would say you are never
2:32allowed to touch soda
2:33ever right and we just watch them for 20
2:3530 years
2:37and then another group we would force
2:39them to drink a lot of soda
2:40right you have to have a two liter
2:43bottle of soda
2:44every single day for the next 20 years
2:47right
2:47and then we would watch their health um
2:50and we would compare
2:52the the results of the group of people
2:54that never ever ever ever touched the
2:56soda ever
2:57and the group of people that were forced
2:59to guzzle down a 2-liter bottle of soda
3:02every single day for 20 years
3:04and see if there's any problems if
3:06there's any differences between the two
3:09right obviously there are ethical
3:12problems with
3:13doing something like that and logistical
3:14problems all kinds of problems right
3:16people don't want to uh
3:18necessarily volunteer for for something
3:20like that so there's problems
3:22with implementing an experiment um
3:25in many cases um so
3:28we settle for second best and
3:31observational studies are definitely
3:34second best always experiments are
3:36always better
3:37than the the results of an experiment
3:39are always better
3:41than a the results of observational
3:44study
3:45experiments lead to being able to
3:48establish
3:49causality that one thing causes the
3:51other
3:52conclusively that's the results of an
3:54experiment
3:55we can scientifically prove that thing x
3:59causes thing y right we can make that
4:01connection
4:02um that's through a designed experiment
4:05an observational study cannot do that
4:08ever right because in say for example in
4:11the case
4:12of following a group of people and
4:15seeing what their habits are
4:16and deciding oh look we've just kind of
4:19paid attention to these people
4:20and the ones that chose to drink a lot
4:22of soda um
4:24it seems like maybe they have more
4:25medical problems but
4:27that is um that's not going to be a
4:30conclusive
4:31connection between the two right if you
4:33can imagine the soda companies are just
4:35going to
4:36fight back and say well you know maybe
4:38it wasn't the soda maybe
4:39those people also eat a lot of potato
4:41chips maybe it's the potato chips that
4:43are doing it
4:44you know how am i supposed to know that
4:46that's not the case
4:47right so um we we cannot establish
4:51a direct causality relationship between
4:55a uh explanatory variable and a response
4:58variable we cannot do that with
5:00observational studies
5:01but sometimes we have no choice we gotta
5:04we gotta do what we can
5:05um and for various reasons could be
5:07money could be ethical reasons
5:10we can't do experiments all the time we
5:12just have to
5:13kind of observe what happens okay
5:18are there any questions between the
5:19differences between an observational
5:21study
5:22versus a designed experiment
5:30no no no it's a lot of ethical stuff
5:33that happens
5:34uh with designed experiments um
5:37during the uh you know late 1930s
5:41to early 1940s uh
5:44there were a lot of medical experiments
5:47actual experiments
5:48done on people uh by the
5:51nazis in in germany
5:54and they were good scientific
5:56experiments so they
5:58they produced good valuable data
6:02but they were done in very cruel and
6:05inhumane ways
6:06um you know the the the experiments
6:10would be
6:10maybe to explore what the brain does for
6:12example
6:14so they would take a perfectly healthy
6:16person
6:17and operate on their brain and like
6:19remove a chunk of their brain
6:21and then close them back up and see what
6:23happens oh look now he can't speak
6:26okay so that's how that brain that part
6:28of the brain controls
6:29uh speech you know really cruel cruel
6:33experiments
6:34like that and we did gain a lot of
6:36useful scientific knowledge
6:38uh but at a very very high cost very
6:41very cruel
6:42cost and after the war there was there
6:45was a lot of debate about
6:47whether we should use that data whether
6:50we should
6:50accept it and then you know well you
6:53know the damage has been done
6:55and there you know there was nothing you
6:57could do after the fact
6:59uh so some people said you know we
7:01should at least
7:02honor their memories of all the people
7:04that were you know
7:05sacrificed and all the all the pain that
7:08that people
7:09were were inflicted uh that was
7:12inflicted on the people and at least
7:14use the data but others think that
7:17you know if we use the data then we give
7:21license to other people
7:22in the future to do the same thing
7:24because they know that oh they'll
7:26they'll judge me poorly at the moment
7:28but whatever results i get
7:30they'll eventually use them um so
7:34i don't know i don't know what you guys
7:35think should we have used that
7:37information that data because you know
7:40the damage is done so we might as well
7:42reap some reward out of it you know all
7:45those people that suffered
7:47or should we reject all that information
7:49and data
7:50because it was obtained in such a such a
7:52terrible way
7:53uh so that we don't incentivize people
7:56in the future for
7:58doing the same thing
8:01so who knows something to think about
8:03there's a lot of great stuff about that
8:05if you want to google
8:06uh information about uh nazi experiments
8:10um uh during you know mostly in
8:12concentration camps and
8:14stuff during during world war ii um
8:17and you know that's a lot of a lot of a
8:19lot of very interesting
8:20information there anyway so no questions
8:24between the difference between an
8:25observational study versus a designed
8:27experiment
8:32no no okay so
8:36this is the gold standard designed
8:37experiment if you can do it right
8:40if there is no moral dilemmas if money
8:43is not a problem
8:45um if you have access to whatever it is
8:48you're doing you know the experiment
8:50experimenting on um you know sometimes
8:53just access is the problem like
8:55maybe you want to do an experiment on
8:59i don't know martian rocks
9:03and we have a very limited supply of
9:05them so
9:07maybe you aren't able to do a full-on
9:10designed experiment
9:11um or maybe the experiment involves
9:14being in mars
9:15uh to really do the experiment so
9:18sometimes you know you have no
9:19you don't have access to the subject or
9:21the material
9:23uh that you want to do an experiment on
9:26sometimes it's just really expensive
9:29right maybe you don't have the funds
9:31to do that kind of a thing um
9:34for example um
9:36[Music]
9:39when a car company wants to bring an
9:42automobile
9:43to to production um
9:46one of the things they have to do is
9:47they have to subject
9:49some of their cars to accidents so they
9:52have to like
9:53give away a bunch of cars to government
9:55agencies
9:56so they can crash them right hit them
9:58from the side hit them from the front
10:00hit them from the back
10:02um so it's pretty expensive
10:05for the company to get one of their
10:08vehicles to be
10:09certified that's why a lot of cars are
10:13not road worthy they're not legal
10:16to to drive on the streets not
10:18necessarily because they're not good
10:20vehicles
10:21um it's just that the company never
10:24decided to get them tested that way
10:26right think of like really expensive
10:29hyper cars that are
10:31you know three million dollars for the
10:33car
10:34um they might not be legal on the
10:38streets
10:39not because they're not safe cars or
10:41anything but because the company didn't
10:43feel
10:43like handing over five or six three
10:46million dollar cars to the government so
10:48the government can crash them
10:50and certify that they are road worthy
10:52right so the experiment is just way too
10:54expensive
10:55um and they chose not to do it
10:59so design experiment if you can do it is
11:01the gold standard
11:02and that's the only thing that will
11:04establish causality
11:06one thing causes the other period that's
11:09the only thing that will do that
11:11in many many many many things
11:14we just settle for second best which is
11:17an observational study
11:19but at best an observational study can
11:22establish
11:22a connection a correlation
11:26some sort of influence between one thing
11:29and another
11:30but it could never establish that one
11:32thing causes the other
11:33it could not establish causality right
11:36that's that's one of the big things in
11:37pop culture
11:38that people hear uh that you know some
11:42sort of study
11:43made a connection between two things and
11:45they jump to the conclusion that that
11:47must mean causality one thing causes the
11:49other
11:50and that's just not true um
11:53at best it establishes a correlation a a
11:56relationship between the two things that
11:58there's some
11:59similarities there or that some
12:02sort of interaction is happening between
12:05the
12:06response variable and the explanatory
12:08variable but we cannot establish
12:10causality
12:13okay we're good questions at all about
12:16observational studies versus designed
12:18experiments no no okay now within the
12:22world of
12:22observational studies we have a couple
12:25of flavors here we have cross-sectional
12:27studies
12:28we have case control studies and we have
12:30cohort studies
12:31so cross-sectional studies so are
12:33observational studies that collect
12:35information about
12:36individuals at a specific point in time
12:39or over a very short period of time
12:42this could be as simple as a survey you
12:45just
12:45stand in a corner somewhere and hand out
12:47surveys
12:48and you get gathering information from
12:50them um
12:52so that could be a cross-sectional study
12:54or maybe you
12:55you obtain information in the form of
12:58i don't know maybe you way people you
13:01you
13:02you go to a intersection with lots of
13:05people are walking by and you kind of
13:07randomly pick some people when you weigh
13:08them
13:09and there you that's how you're
13:10gathering your data um
13:12or and asking them some questions maybe
13:14about their eating habits or
13:16how much they what they've eaten that
13:18day maybe you just want to get
13:19information
13:20about what kind of a breakfast the
13:22typical american has
13:24um or you know how much sleep they had
13:27maybe you stopped people in the morning
13:29at a starbucks and you asked them how
13:30many hours of sleep did you have last
13:32night
13:32and you know you might get oh three
13:34hours uh seven hours
13:36you know i haven't slept no that kind of
13:38information cross-sectional
13:40uh so you collect your data about the
13:42individuals at a very specific point in
13:44time
13:44um at a specific point in time or over a
13:47very short period of time
13:48a case control study these studies are
13:50retrospective meaning
13:52that they require individuals to look
13:54back in time or require the researcher
13:56to look
13:56at existing records in case control
13:59studies individuals who have certain
14:00characteristics are matched
14:02with those that do not okay so
14:05i think as the definition uh implies
14:08uh you look back at records you know you
14:11figure out
14:12something about birth rates in 1970
14:16or something maybe you're maybe you're
14:19localizing it to a particular state
14:22that had some traumatic event i don't
14:24know like let's say for example
14:27we're looking at louisiana uh during the
14:31the katrina hurricane we wanted to see
14:33how that impacted
14:35birth rates uh you know that kind of
14:37thing
14:38did people choose to have children um
14:42less children or more children that year
14:44was it impacted at all
14:46right okay uh cohort studies a cohort
14:49study
14:50uh first identifies a group of
14:52individuals to participate in the study
14:54the cohort the cohort is then observed
14:56over a long period of time
14:58over this time characteristics about the
14:59individuals are reported
15:01because the data is collected over time
15:03cohort studies are prospective right so
15:05this is kind of like that soda study i
15:07said
15:07uh let's just uh watch a thousand people
15:10for the next 20 years
15:12and keep track of their medical uh
15:15activity
15:16right so we at the beginning take a good
15:18detailed
15:20uh medical record of where they stand at
15:23that
15:24starting point and then maybe once a
15:26year we
15:27invite them to come in and keep track of
15:30how they're doing
15:31um and also get information about what
15:34their soda consumption has been like
15:36right it's an observational study so
15:37we're not forcing anything
15:39maybe they choose to drink a lot of soda
15:41on their own or maybe they choose to
15:43stop drinking soda on their own
15:45right we don't we don't have any and we
15:47don't influence the
15:48uh the explanatory variable
15:51in any way in in any of these
15:54observational studies
15:55we just observe what they're doing and
15:57record
15:59good once we have all that data then of
16:02course we try and
16:03uh see if there are correlations
16:05connections between
16:07the explanatory variable which in this
16:09case in my little scenario would be
16:12drinking of soda habits and the
16:16response variable which would be health
16:18maybe
16:19in particular i might be looking at
16:20diabetes
16:24or might be weight maybe soda
16:26consumption and weight
16:28obesity or diabetes
16:32who knows right there might be heart
16:34issues all kinds of things
16:36um maybe parkinson's or you know
16:39dementia
16:40maybe i'm looking to see if people that
16:42drink a lot of soda
16:43once they are older are they more likely
16:46to develop alzheimer's disease or
16:48something
16:50good any questions about those here's
16:54some examples
16:58determine um let's determine whether
17:01each of the following studies depicts an
17:03observational study or an experiment if
17:05the researcher concluded an option
17:07if the researchers uh if the researchers
17:10conducted an observation study
17:12determined the type of observational
17:13study
17:14okay researchers wanted to assess the
17:16long-term psychological effects of
17:17children
17:18evacuated during world war ii they
17:20obtained a sample of 169 former
17:23former evacuees and a control group of
17:2643 people
17:27who were children during the war but
17:28were not evacuated
17:30the subject's mental state were that was
17:32were evaluated using questionnaires
17:35uh it was determined that psychological
17:37well-being of individuals was
17:39adversely affected by evacuations right
17:42so this is
17:43an observational study right we did not
17:45do an experiment we not
17:47separate people into two groups and then
17:51uh made it made some sort of decision
17:54about how to impact them right
17:56if we wanted to convert this into an
17:57experiment we would say back
18:00during world war ii we grabbed two
18:02groups of
18:03people children in this case randomly
18:06assigned them to two groups
18:07and then we randomly chose one group to
18:09be evacuated and the
18:11other group to not be evacuated right so
18:14that would be an experiment but this is
18:17not an experiment
18:18uh we have no control the researchers
18:21the uh
18:21looking at this had no control over who
18:23got evacuated and who didn't get
18:24evacuated
18:27and this is a case case control um
18:29observational study
18:33right case control right the studies are
18:37retrospective we're looking back at
18:38records
18:39back to world war ii and determining
18:42what happened
18:46okay um another
18:50another uh example xylitol
18:54has proven effective in preventing uh
18:56dental
18:58dental cavities carries dental cavities
19:02when included in food or gum a total of
19:0575
19:06peruvian children were given milk uh
19:09with with and without xylitol and we're
19:12asked to evaluate
19:13the the taste of each uh overall the
19:16children prefer the milk flavored with
19:18xylitol okay so this is an experiment
19:21we're
19:22affecting the uh we're impacting
19:25what they do right we're intentionally
19:27giving some children this drug and some
19:29children
19:30we're not giving them this drug um
19:33there's no there's no ethical issues
19:36here
19:36hopefully we're very confident that this
19:39drug will not have
19:40a negative effect on people so um
19:44we feel ethically secure that we're
19:47allowed to give it to people
19:49um and so this is an experiment
19:52good looks like all we were really doing
19:54though is evaluating whether the
19:55children
19:56preferred the milk uh with xylitol
19:59or without xylitol right so i guess
20:02flavor
20:03is is what we're really checking here
20:06um a good follow-up would be to
20:09come back and follow them some number of
20:12years later and see
20:14what their dental health was like
20:17assuming we continue with this treatment
20:20so it's a designed experiment
20:24good a total of 974 homeless women in
20:27los angeles
20:30area were surveyed to determine their
20:31level of satisfaction with the health
20:33care provided by the shelter clinics
20:35versus the health care provided by
20:36government clinics the women
20:39reported greater quality satisfaction
20:41with the shelter
20:43and outreach clinics compared to the
20:44government clinics okay
20:46so this is an observational study right
20:49we're not randomly assigning
20:51some women to go to one clinic and some
20:52women to go to another clinic they're
20:54choosing for themselves
20:56which clinic they uh they would like to
20:58go to and when they would like to go
21:01um and they are just reporting
21:05what their opinion is on on on the
21:07clinics so it's definitely an
21:08observational study
21:10and it's also a cross-sectional kind of
21:13study
21:14right so we're it's a observation of
21:16study that collects information about
21:17individuals at a specific point in time
21:20right we're just stopping them at that
21:22point in time
21:23and asking them what they think uh in
21:26that point in time
21:32okay another example uh
21:36the cancer prevention study two
21:39uh is funded and conducted by the
21:41american cancer society its goal is to
21:43examine
21:44the relationship among environmental and
21:45lifestyle factors on cancer cases
21:48by tracking approximately 1.2 million
21:50men and women
21:52study participants completed an initial
21:54study questionnaire in 1982
21:56providing information on a range of
21:58lifestyle factors such as diet
22:00alcohol and tobacco use occupation
22:03mental history and family
22:05cancer history these
22:09these data have have been examined
22:11extensively in
22:12relation to cancer mortality vital
22:15statistics of study participants
22:17is updated biannually uh causes of death
22:21has been
22:21documented for over 98 of the deaths
22:24that have occurred
22:26uh mortality mortality follow-up to the
22:30cps two participants is complete through
22:332002 and is expected to continue for
22:36many years
22:38okay so they're just keeping track of
22:40these people
22:42right so this is a cohort study it's
22:44over a long period of time you grab the
22:46big group of people
22:48you kept track of their medical data and
22:51uh on a bi-annual basis you go back and
22:54get
22:55more data keep track of these people
22:57until they die
22:58uh and then they they gather cause of
23:01death
23:01information uh so that you can
23:05try and see if there's a connection
23:07between their
23:08lifestyle factors and cancer
23:11and cause of death
23:14any questions and questions any
23:15questions
23:19[Music]
23:21okay more definitions um a sentence is a
23:24list of individuals in a population
23:27along with certain characteristics of
23:28each individual
23:30okay a census is when you have
23:33um information a particular information
23:37about a particular characteristic
23:39from every individual in your population
23:43um at the moment we're going through the
23:45us census
23:47uh right so written into our
23:49constitution
23:51uh it is important that every 10 years
23:53the government counts
23:55how many people live in the country um
23:58and where do they live which state do
24:00they live in
24:01right so the the the government is
24:03charged with doing that every 10 years
24:05it's built into our constitution
24:07uh and we're doing it this year right uh
24:10so
24:11that's a census um don't confuse that
24:13with
24:14samples right a sample is a little
24:16subgroup of the population
24:22okay um simple random sampling
24:26so our learning objective in this next
24:28section is just to learn about samples
24:30right
24:30to compare that contrast that with a
24:33with taking a census
24:37okay so random sampling is the process
24:40of using
24:40chance to select individuals from a
24:42population to be included in a sample
24:46if convenience is used to obtain a
24:48sample the results of the survey
24:49are meaningless right convenience or
24:53members that self-select right so
24:56if you ask for volunteers hey who wants
25:00to be a
25:00volunteer in this sample then
25:04it's kind of mathematically
25:06statistically useless
25:07information good those
25:11members of the sample that come from the
25:14population
25:15must be randomly selected so that you
25:17have a good
25:18chance of of having a sample
25:21that accurately reflects the population
25:25uh by statistical characteristics right
25:28so for example
25:29if i wanted to get a sample of 100
25:33students
25:34that represent the entire population of
25:37our college
25:38right our college has nearly 20 000
25:41students
25:42so if i just wanted to get 100 students
25:45that do a pretty good job of reflecting
25:48the college
25:49right in terms of ages and race
25:53and who works full-time who doesn't work
25:55full-time
25:56children live at home don't live at home
26:01long commute short commute they
26:04learn they take online classes versus in
26:07person
26:08right all those characteristics of the
26:10population of the entire school
26:12i want to be able to have a small sample
26:15that reflects that so that if i
26:17study my little sample right it's really
26:19difficult to study
26:20the entire college 20 000 students
26:24if i wanted to learn how the
26:27the students feel about something right
26:30is there enough parking at our school
26:32is is that a problem how do i find that
26:35out right i could ask
26:36every single student at the college but
26:39that's a lot of work that's a lot of
26:40students
26:41so instead what we want to do is get a
26:43little sample
26:44that does a really good job of
26:46impersonating
26:47the entire college it's a little
26:49microcosm of the entire college right
26:51there in that tiny little group of 100
26:53people
26:54and so that 100 people should do a good
26:57job of reflecting
26:58um the number of men versus the number
27:01of women in the entire college right so
27:03if in the entire college we're 60
27:05percent female
27:06then my sample should be about 60 female
27:09right
27:09approximately um if my entire college
27:13is 32 asian
27:17uh then my sample should also be about
27:20that 32
27:21right um if my entire population
27:25uh 20 of the entire population has
27:28children
27:29then the same should be of my little
27:31sample ideally right should be like a
27:33little microcosm
27:34so that when i ask questions to those
27:37people in my sample those 100 people and
27:40i get their
27:42perspective on things there's a good
27:44chance
27:45though not guaranteed there's a good
27:46chance that those 100 people
27:49will do a good job of representing the
27:52entire school and how the entire school
27:54feels
27:54about certain things good but that's why
27:58it's important for those
27:59100 people in my sample to be selected
28:02at random from the population because if
28:05i just ask for volunteers
28:07for example then there's a very good
28:10chance
28:10that um that my little sample
28:14won't really reflect the entire school
28:17right there might be for example
28:20a a big chunk of students that are very
28:23shy at my school the entire college
28:25there's thou there's a couple thousand
28:27students that are really
28:28really shy so therefore just by
28:31definition
28:32they wouldn't volunteer um for a
28:36first for uh to be in a sample and so
28:38therefore
28:39when i'm looking at my sample my sample
28:42doesn't really reflect
28:43that group of people that are really shy
28:45right these people are really vocal and
28:47active
28:48and they chose to be in that sample so
28:51you know they're the loudmouths they're
28:53the the
28:54the extroverts that are very vocal
28:58about how they feel and that might not
29:00be
29:02truly reflective about how the entire
29:04college feels
29:05well that's one example uh two
29:09maybe if i ask for volunteers
29:12it's more likely that people that have a
29:14lot of free time decide to volunteer
29:16you know they're not doing anything
29:17anyway they're bored okay sure i'll
29:19participate why not
29:20people that are really busy they have a
29:22full-time job they have
29:24a full-time school load they have three
29:26children at home
29:28like they're just busy um the idea of
29:31having to participate
29:33volunteering to participate in some some
29:36survey thing it doesn't appeal to them
29:39so they kind of pass on it and so you
29:41end up having a sample
29:43that doesn't really have those really
29:45really busy people
29:46in that sample so again our sample fails
29:49to really
29:50truly reflect the entire population good
29:54so if you don't have a random sample
29:57it's not really valid okay a lot of it
30:00is convenience
30:01you know um for example any time you
30:04ever have
30:05a uh you know like an internet survey of
30:08some kind
30:09um you know that is laziness
30:13from from the people doing it uh they
30:16just
30:16you know we we wanted to get information
30:19about the
30:19uh the the school the uh student
30:23population
30:24so i could just send out an email to 20
30:26000 students
30:27and then just take the ones that
30:30responded
30:31right that's that's me being very lazy
30:34sure
30:34i sent out 20 000 emails maybe i get 100
30:37back
30:38there's my sample but again that doesn't
30:40really
30:41mean that that sample is going to do a
30:43good job of reflecting the entire
30:46school population i mean there might be
30:48a group of people
30:50that are very bad with technology
30:53and they don't get emails and there's
30:56still people out there that
30:57have trouble uh with computers and
31:00logging in
31:01and figuring out what their email is um
31:05so if my method of trying to get a
31:07sample
31:08of students is just to send out an email
31:10and see who responds
31:12um then my sample isn't really gonna
31:15include people that are
31:17[Music]
31:19technology deficit right
31:22obviously um also um
31:26maybe people have blockers in place
31:30and they see my email and they just
31:32think it's spam
31:33and like delete it um so again
31:37my sample is gonna it's gonna fail
31:40to include those kinds of people into my
31:43sample
31:44and ultimately what you get is a sample
31:47that doesn't really reflect the
31:48population
31:50and if your sample doesn't really
31:51reflect the population
31:53then asking questions to that sample
31:56doesn't mean that you're going to get
31:58responses that are reflective of the
31:59entire
32:00population taking measurements of
32:02anybody in that population doesn't
32:04necessarily mean
32:06that you're getting useful information
32:08that's reflective
32:09of what the entire population
32:13uh is reflective of the entire
32:16population
32:18okay so am i making it crystal clear
32:20that it's incredibly incredibly
32:22incredibly important
32:24that you do random sampling
32:27for your sample if it's not random it's
32:30junk
32:32useless junk right that's why any kind
32:36of internet survey
32:38where people just volunteered to submit
32:40their results
32:41it's all junk right there's absolutely
32:44no statistical validity
32:45to that good good
32:50add to it that um say for example just
32:54surveys you know how do you feel about
32:56thing x and then you know that's out
32:57there when you go to some website and
32:59it's keeping track of who said yes and
33:00who said no
33:01uh to to to a particular thing it's like
33:04a hot button thing
33:06um you know keep in mind that a lot of
33:09people
33:11when you throw something like that out
33:12there you're gonna get a lot of
33:13responses from people that are
33:15very vocal about that issue uh very
33:17passionate about that issue in one way
33:19or the other
33:20um and so that again might not be
33:24reflective of how the entire population
33:26feels
33:28that's just it potentially and might me
33:30might be a very small
33:32group of the population that's very
33:36vocal
33:38questions questions like for example
33:42let me throw something out there it's
33:43not too controversial um
33:46in my neighborhood uh there is
33:50a part of the neighborhood that has a
33:52fantastic
33:53view of the ocean uh so they are
33:57very passionate about laws that
34:00impact their their ocean view
34:03so there's a lot of laws that involve
34:06like how tall trees can be um
34:10fences how tall can your neighbor put up
34:12a fence
34:14you know stuff like that they're very
34:15very uh involved when it comes to
34:18anything
34:18that impacts their their view so
34:21if a new law comes up that says you know
34:24what we should let people plant whatever
34:26tree they want whenever they want no
34:28matter how tall it is
34:29um and maybe the the city wants to put
34:32out a survey
34:33to see how people feel about heights of
34:35trees
34:36uh i i'm i'm guessing you're gonna get
34:40a very very visceral response
34:43from those people that impacts them
34:47so if you're just looking at the survey
34:48you know wow we put the survey out and
34:51it was clear from the survey results all
34:55the people that volunteered there to
34:57participate in the survey
34:58i think you're going to get a lot of
35:00really strong
35:02negative don't do it kind of things so
35:04if you base yourself just on that you
35:06think wow the entire city is very
35:08passionate about that
35:09that that particular law they definitely
35:11don't want tall trees
35:13but that's not really true i think most
35:16of the city
35:17just doesn't really care one way or the
35:19other or maybe they actually do want
35:21tall trees you know
35:22away from the parts where you get really
35:24nice views of the ocean
35:26away from there people might like nice
35:28tall
35:29beautiful fruit trees so maybe they they
35:32like that
35:32so again my point being that in in this
35:35city
35:36a very small group let's just say 10
35:38percent of the people
35:39really really really care about having
35:42no tall trees so in a
35:46volunteer survey they would probably
35:49make themselves vocal and throw
35:51themselves into that survey and heavily
35:53push
35:54the conversation toward short trees only
35:56no tall trees
35:58so some outsider that's looking at just
36:00this survey
36:02might get the wrong impression about how
36:04the entire population feels
36:06good so that's why random sampling is
36:10important
36:11right you can't let the people the
36:13members of the population
36:15self-select and choose to participate in
36:18it
36:18you're not going to get a sample that
36:20truly reflects the population
36:22now having said that just like before
36:25sometimes you have
36:26no choice um there might be uh
36:29you know logistical reasons there might
36:31be money reasons
36:33there could be various reasons for why
36:35you can't have
36:36a truly random sample so we might have
36:40to settle
36:41for a second best a almost random
36:44sampling
36:45uh or just give up on the ram random
36:48sampling altogether
36:49um and then just recognize that this
36:53sample is not random
36:54so therefore all the conclusions are not
36:58as scientifically secure as they could
37:00be if we had a random sampling
37:03oh but like i said sometimes you can't
37:05you can't uh
37:06you can't do better you might have to
37:08settle for something right but the gold
37:10standard
37:11is random sampling any questions about
37:14anything that i just
37:15rambled on about
37:21okay within the world of random sampling
37:24um there is a few types of sampling
37:27techniques that we're going to look at
37:29uh so the first thing we want to look at
37:31is simple random sampling
37:33so a sample of size n from a population
37:36of size capital n is obtained through
37:38simple random sampling
37:40if every possible um
37:43sample of size lowercase n has an
37:46equally likely chance of occurring
37:49the sample is then called simple random
37:51sample okay this
37:52is the classic gold standard simple
37:54random sampling
37:56and what i want you to visualize uh for
37:59this kind of sampling
38:00is that you know imagine that i need to
38:03take
38:04five students from our class simple
38:06random sampling
38:08i want to create a method so that
38:11any five students uh have an equally cha
38:15equal chance of being selected into into
38:18this
38:18sample so the most classic way to do
38:22that and visualize it
38:23is everybody's name right just write
38:25everybody's name on a little piece of
38:27paper
38:28you know and then fold up each little
38:30piece of paper throw it into a big hat
38:33shake it up and i'm just gonna draw five
38:36names
38:36one at a time i'm never gonna repeat the
38:39same name right once i
38:40pick out one name you know reach in
38:42there tom you're in my
38:44sample right i'm not going to put tom's
38:46name back in the hat
38:48so you reach in there you grab one name
38:50you reach in there you grab another name
38:51or you do that five times you got five
38:53people there you go
38:55okay so that's that that technique will
38:58allow
38:59any five students uh in the in the
39:02population from being selected
39:04right so there's i'm not limiting any
39:06combination of uh
39:07of students any combination is is
39:10allowed um and they're all equally
39:13likely of occurring
39:15right equally likely occurring from the
39:17stance of
39:18at the beginning before you've done
39:20anything then
39:22any combination is equally likely of
39:24occurring
39:25once you get through the middle of the
39:27process and you start
39:28singling out some of the people then
39:30some combinations become less and less
39:32and less likely
39:33in fact some of them become impossible
39:35once you start making some choices
39:37so uh when i say that they're all
39:40equally likely that's
39:41you know when you're before you start
39:42selecting any names and you're thinking
39:44about how you're going to do this
39:45process
39:46all uh all of them are equally likely of
39:50occurring
39:51good
39:54okay so names in a hat that's the gold
39:57standard that's simple random sampling
39:59so here's just a quick example uh
40:02illustrating that suppose that a study
40:03consists of five students so here is the
40:07entire population is five students
40:09bob patricia mike jan and maria
40:12so two of the students must go to the
40:14board to demonstrate homework problems
40:15list all possible samples of size two
40:18okay so it could be bob and patricia
40:21bob and mike bob and jan uh
40:25bob and maria or it could be
40:28patricia and mike patricia and jan
40:33patricia and maria and then mike and jan
40:37mike and maria or then jan and maria so
40:40this is where order does not matter
40:42so we are not including the case of
40:44patricia
40:45then bob right those two are considered
40:48to be the same thing
40:50so here's all possible outcomes there's
40:53one two
40:54three four five six seven eight
40:57nine ten possible outcomes
41:04okay um so if all 10
41:07outcomes are equally likely of occurring
41:10um
41:11then we have ourselves a simple random
41:13sample
41:16so obtain a a frame that lists
41:20all the individuals in the population of
41:22interests number individuals
41:24in the frame
41:27what steps for obtaining a simple random
41:29sample obtain a frame that lists all the
41:31individuals in the population of
41:32interest
41:33number the individuals in the in the in
41:35the frame from one through capital n
41:38uh use a a random number table or
41:41graphing calculator or statistical
41:42software to randomly generate
41:44and numbers where n is the number of
41:46desired samples
41:47okay so we can use technology to do that
41:50as well
41:50so um the classic visualization is to
41:54put
41:55names right put bob's name a little uh a
41:57little piece of paper fold it up throw
41:59in the hat
41:59the patricia's name and little piece of
42:01paper fold it up throw it in the hat
42:03right so we can do that we'd have five
42:05little pieces of paper
42:06shake it up randomly pick two you know
42:09that's one way to do it
42:10the other way to do it that this is
42:11suggesting is to use technology and as
42:14as um when we only have five people
42:18what i suggested with the names is
42:19probably the easiest thing to do
42:21but this method the advantage of this
42:24method is that
42:25if we have a really big list like
42:28hundreds or thousands of people it would
42:30be a lot of work
42:31to write down everybody's name on a
42:34little piece of paper
42:37and then randomly pick the values right
42:40so instead it would make a lot more
42:41sense
42:42to use technology
42:45jan so this is a very popular way to do
42:48it
42:49maria okay capital n represents the
42:53number of people in our population
42:57number of
43:02people in population
43:11whereas lowercase n is the number of
43:14people in our sample
43:17number of people
43:21in sample now i say people in this case
43:25because we're
43:26talking about people but um
43:29in many cases this might be you know car
43:32companies or soda brands
43:33or you know types of animals right so it
43:37could be anything
43:38but in this case it happens to be people
43:40so it could be number of objects
43:41in our population okay so number of
43:45people in the population
43:46um sorry in the population versus
43:49the sample lower case n is the sample
43:52okay so one thing we could do is
43:54note that n is
43:57goes from one through five in this
43:59example and so we can just have
44:01technology
44:02randomly give us some numbers right so
44:04randomly give me two numbers
44:06and so you go to a computer program or
44:08something and say
44:09between one through five randomly give
44:11me two numbers so it might give you the
44:12number five and the number two for
44:14example
44:15and then you can come back and go oh
44:16that means it's maria and patricia
44:19remember order doesn't matter
44:22so that's another way to do simple
44:24random sampling
44:27and when n is really big so instead of 5
44:29maybe there's 5 000 names
44:31we don't want to do a little piece of
44:33paper with 5 000 people
44:35so a computer program uh does a better
44:37job of
44:39of repeating the same process
44:44good questions questions questions
44:46questions
44:47uh we will not be doing that in our
44:50class and you know in practicality we
44:52will not be really doing that at all
44:54so just getting the visual of what
44:56simple random sampling is
44:58is what's important for us and so that
45:00visual of writing everybody's name
45:02putting it in a hat um and drawing names
45:05is what you should think of when you're
45:07thinking of what is a
45:08simple random sample right all possible
45:11outcomes
45:12can happen nothing's being restricted
45:15and they're all equally likely of
45:16occurring
45:18um an example of a simple random sample
45:21in the 112th
45:23congress of the united states had
45:27of the united states had 435 members in
45:29the house of representatives
45:30explain how to conduct a simple random
45:32sample of five members to attend
45:35a presidential luncheon then obtain the
45:37sample okay so put the members in
45:39alphabetical order
45:41um numbers from 1 to 435
45:46and then uh randomly select five
45:49right use a randomly select five numbers
45:52using a random number generator
45:54okay so exactly what we've we've
45:58explained
46:04so the random number generator will just
46:07spit out some numbers for you 182
46:10207 409.
46:13so randomly assigned numbers and then
46:15you go back and you match up
46:17the number to the name of the person
46:20and that's that's how you can get your
46:22sample right but again
46:24visualize all one 435 names
46:28are written down a little piece of paper
46:30throw them in a hat shake them up
46:31pick five people that simple random
46:33sampler
46:36there are a few other types of um
46:39sampling methods we're going to look at
46:40one of them is stratified sampling the
46:42other one is
46:44systematic sampling and the other one is
46:46cluster sampling
46:49so a stratified sample is obtained by
46:52separating the population into
46:54non-overlapping groups called
46:56strata and then obtaining a simple
46:59random sample
46:59from each strand the individuals within
47:02each strand
47:03should be homogeneous or similar
47:06in some way so for
47:10example what if i wanted to get a sample
47:13of students
47:15um i don't know you're gonna go maybe
47:18there's a
47:19maybe uh there's a tv show and they
47:22wanna
47:23you know have six students represent el
47:27camino
47:27college so they want six students so if
47:30we
47:31if we take all twenty thousand el camino
47:34college students
47:35and we put all their names in a hat and
47:37shake it up and we're just going to pick
47:39six students
47:40then any six students are equally likely
47:42of being selected
47:44and so there is that simple random
47:47sampling
47:48on the other hand um maybe the school
47:52specifically wants to make sure that
47:55there are
47:55three males and three females i don't
47:58know why they just
47:59that's what they want that's the that's
48:01the representation that they want
48:03uh so um they don't they wanna
48:06they don't wanna allow the possibility
48:09that some other
48:10uh gender makeup
48:13goes toward the group they wanna make
48:15sure it's three males three females
48:17so what they could do is separate the
48:19names
48:20of all the males in one camp and from
48:23that group
48:24take three and then separate all the
48:26females and then from that group
48:28take three right so there you can
48:30generate your sample of six
48:32students guaranteeing that three of them
48:35are male and three of them are female
48:37good so before you make your your
48:39selection
48:41the act of separating all the males and
48:43all the females
48:44that's what's the strata thing okay
48:48strata and then you've taken a sample
48:50another thing is
48:51maybe we want to uh create a panel
48:55of students to to talk about their
48:57experience at
48:58el camino um but maybe what we want to
49:01do
49:02is get one student from
49:05each um from each degree that we offer
49:09right so we don't want to get a panel
49:12that is
49:13heavily represented in one field versus
49:16another
49:17we want one business major we want one
49:19math major
49:20one uh engineering major one
49:24you know biology major we want one from
49:27each
49:27each major that we offer so
49:31you know we use computers to look at the
49:34uh
49:34you know the students and we breaking up
49:37break them down
49:38by strata right based on what their
49:41uh declared major is and then once we
49:44have them broken down in strata declared
49:47major
49:48then we go in there and we randomly pick
49:50one student from each major
49:52thereby guaranteeing that we'll end up
49:54getting a sample
49:56that has the particular makeup that we
49:58want to
49:59have good do we see how that's not
50:02simple random sampling
50:03because in simple random sampling any
50:07any makeup of students is equally likely
50:10of occurring so if i just leave it up to
50:13simple random sampling
50:14it's possible that my sample is
50:18just is made up of students that come
50:21from one field
50:22that is overrepresented by students that
50:24have from one from one major
50:27right like let's say maybe our school
50:30has a lot of business majors that's just
50:34heavily heavily in you know impacted the
50:37the program is heavily impacted there's
50:39a lot of business majors
50:40so if we just randomly select 100
50:42students
50:43there's likely going to be a lot of
50:45business majors in that sample
50:48which on the one hand is a good
50:50representation of the school the sample
50:52might do a good job of
50:54accurately reflecting the the makeup of
50:57the school so it might be a good sample
50:59from that perspective
51:01but if you intentionally
51:04want to make sure that each student from
51:07each major is chosen um
51:10then you might want to intervene and
51:13create a sample
51:15in this method are there any questions
51:26okay so here's an example in 2008 the
51:28united states senate had 47 republicans
51:3151 democrats and two independents the
51:34president wants to have a luncheon with
51:35four republicans
51:37four democrats and one other obtain a
51:40stratified sample in order to select
51:42members who will attend the luncheon
51:44okay so we don't want to just randomly
51:45select um looks like
51:47the president is going to have lunch
51:49with nine people
51:50four four and one so if you just say any
51:54nine people at random then you are
51:57allowing the possibility that that
51:59sample of nine people
52:01has a different makeup than this right a
52:03president is asking for four republicans
52:05four democrats
52:06one other good so a simple random
52:09sampling
52:10would not guarantee that for you simple
52:12random sampling
52:13would say any uh outcome is equally
52:16likely of occurring
52:18so um
52:19[Music]
52:21there are many outcomes that would
52:23result
52:24that would not have this makeup right
52:26there's some outcomes where it'd be
52:27all democrats or all republicans there'd
52:30be some
52:31outcomes that have none of the
52:33independence
52:34probably a lot of outcomes that have
52:35none of the independence since there's
52:36only two of them
52:39right in this scenario who's the most
52:43uh influential uh group
52:46right some could argue that the
52:48independents are the most influential
52:50group
52:51since there's only two of them and one
52:54of them is going to get to go
52:56to this very important luncheon with the
52:58president
52:59right so if you're one of these guys uh
53:02uh you have one in two chance
53:0450 chance of being selected to go to
53:07this
53:07luncheon whereas if you are either a
53:10republican or a democrat
53:12um the odds of you being one of the four
53:15people from each camp
53:17to be selected are much smaller right so
53:20in a way you could say that these
53:22these four people are more powerful more
53:25influential
53:26um right in a way but there's
53:29lots of ways to look at it anyway so
53:31it's clear that this is stratified
53:33sampling
53:34you're going to take the group of
53:35republicans the group of democrats the
53:37group independents you separate them
53:39by their political affiliation so those
53:42are the strands
53:44and then from each strand you're gonna
53:47randomly choose
53:49the number of people to make up your
53:51your sample
53:56okay moving on the next one is called
53:58systematic a systematic sample
54:00a systematic sample is obtained by
54:02selecting every kth
54:04individual from the population um the
54:06first individual selected at random
54:10is selected at random um
54:13from numbers between 1 and k okay
54:16so um the classic example for this one
54:20is this
54:21right here a quality control engineer
54:23wants to obtain a systematic sample of
54:2625
54:26bottles coming off a filling machine
54:30uh to verify that the machine is working
54:31properly design a sample
54:33a sampling technique that can be used to
54:35obtain a sample of 25 bottles right so
54:37if you can imagine
54:39one of these bottling machines i'm sure
54:41you've seen them somewhere you can
54:42google them
54:43um they're just this conveyor belt
54:46that's just like
54:46flowing out bottles really really fast
54:50right hundreds and hundreds of bottles
54:51are flying by
54:53um and so you want to
54:57randomly uh pick some of those bottles
55:00uh and so if you choose to do it in this
55:03system
55:04systematic method uh what you do is step
55:08one
55:08you're gonna have to choose that first
55:10bottle
55:12and then after you've chosen that first
55:14bottle then you go okay well
55:16after the first bottle selected you know
55:19skip
55:2030 and then pick another bottle and skip
55:22another 30 and
55:23take another bottle for example
55:26good so um
55:30if possible approximate the population
55:32size
55:33so in some cases there will be an
55:35approximate
55:36sample size um maybe it could be
55:41students in one classroom you know
55:42there's about 30 of them
55:44or it could be um people
55:48in a movie theater you know i don't know
55:50the the room
55:51fills about a thousand and it's kind of
55:53a sold out movie so
55:55uh let's say the people in a movie
55:56theater about a thousand people
55:58or okay you could have this um this
56:01scenario of having those bottles kind of
56:03fly by
56:04uh and there's thousands and thousands
56:06like it never it never really ends
56:08uh so you could have that in the case
56:10like that you don't really have a
56:11capital n
56:13okay so determine the the sample size
56:15desired
56:16lowercase n compute n divided by
56:20lowercase n and round down to the
56:21nearest integer
56:23so that gives you an idea approximately
56:26of
56:26how big the k should be um
56:30where k tells you how many you skip good
56:33so now
56:34randomly select a number between 1
56:36through k call this number
56:37p so you're going to start at the ph one
56:41and then do k number of people after
56:44that
56:44so let's let's uh for example if
56:48um in a movie theater
56:56uh with approximately let's say there's
57:00a thousand people
57:03and then when the movie ends they all
57:06single file leave there's only one exit
57:09um and so you can stand outside the room
57:12right so here is like the movie theater
57:15everyone's sitting here watching the
57:18movie having a great time you guys
57:19remember that
57:21so long ago anyway uh
57:24when the movie ends people just kind of
57:28end up single filing out of their way
57:31and so if you stand
57:32right here you can watch them as they go
57:35and then maybe you want to survey them
57:37maybe you want to ask something about
57:39them did they
57:39enjoy the movie um how often do they
57:42come to the movies
57:43did they buy popcorn are they
57:46likely to come back soon you know all
57:49kinds of questions that you can ask them
57:51um and so you want to get a a sample
57:54from them
57:55so let's say let's just follow these
57:56things here so capital n
57:58in my little example here capital n
58:00would be a thousand
58:04you have to choose for yourself
58:05approximately how many people do you
58:07want in your survey
58:08so n is the number of people in my in my
58:11sample
58:12so let's say that i want um
58:16i don't know let's say i want 40 people
58:19approximately 40 people okay so this is
58:22suggesting that in order to find
58:24the k-th value you should take n divide
58:27a capital n divided by lower case n
58:29so k then is equal to
58:32about a thousand divided by about 40.
58:40so that gives you is that 25
59:00yeah 25 i know i look weird for a second
59:03okay anyway um so that's about 25 people
59:07okay randomly select uh randomly select
59:10the number from one through k
59:12so k is that so now we'll let p be some
59:15number
59:15between 1 through 25 1 2
59:183 4
59:2121 22 23
59:2524 25 okay
59:28so there's all the numbers from 1
59:30through 25 and we're gonna let
59:32p be some value in there right kind of
59:35randomly chosen so maybe i let p
59:38be this guy four okay so if we let p
59:42equal to four now we have all the
59:44makings of what we need
59:46so we're saying we're going to stop the
59:47fourth person
59:49and make them a member of my sample hey
59:52let me stop you for a second
59:54i'm going to ask you a bunch of
59:55questions and if you answer them
59:57you know you get a free ticket to come
59:59back to the movies or something
1:00:01okay and then i'm going to count um
1:00:05by k which is 25
1:00:08and then that's the next person i stop
1:00:11so 4
1:00:12plus 25 is going to lead me to 29.
1:00:19so i ask the fourth person oops
1:00:22i'm going to ask the fourth person and
1:00:24then i'm going to ask the 29th person to
1:00:26come out of the room
1:00:27to be a part of my sample and then i'm
1:00:30going to ask
1:00:32the 4 plus twice 25
1:00:37which is going to be equal to 54.
1:00:41then the 54th person is also a member of
1:00:44my survey
1:00:45then it's the four plus triple 25
1:00:51which will be the 79th person
1:00:55right and that's what this is describing
1:00:57dot dot dot dot
1:00:58you just keep going that way this will
1:01:00guarantee
1:01:02that not guaranteed but this will give
1:01:05you approximately
1:01:07the the number of people k approximately
1:01:1025 people will be in your survey
1:01:13right so the first person in my survey
1:01:15the second one the third one
1:01:17the fourth one and so on and so forth
1:01:20any questions about this
1:01:34okay um so
1:01:38this method has its benefits as well
1:01:40sort of
1:01:41you know in this case of trying to stop
1:01:43people coming out of
1:01:45a theater you know this would be
1:01:48a good way to do it because what other
1:01:51way would you do it
1:01:52um i guess you could
1:01:55ask everybody to give you
1:02:01people and go hey if i call your name
1:02:04please don't leave the theater
1:02:06so that you can be part of a survey that
1:02:08seems more messy
1:02:09right people don't want to like you're
1:02:11going to have to take names of a
1:02:13thousand people
1:02:15okay i guess maybe if each individual
1:02:18seat is numbered um
1:02:20then i guess you can put all those
1:02:23numbers
1:02:23into a computer program to generate the
1:02:2625 numbers you want
1:02:28and then maybe at the end of the movie
1:02:30experience you can you can kind of
1:02:32announce it
1:02:33hey if you're seating in seats
1:02:36e25 seats b17
1:02:40right kind of call out the seat numbers
1:02:42that you want
1:02:43and ask people to stay behind so that
1:02:46they can participate in a survey
1:02:49i guess that's somewhat doable
1:02:52um but a convenient
1:02:55way to do it would be to do the
1:02:56systematic sampling if you're kind of on
1:02:58the outside and you're
1:02:59watching people walking out and kind of
1:03:02tackle them down
1:03:05any questions
1:03:11okay next we're going to look at a
1:03:13cluster sampling a cluster sample is
1:03:15obtained by selecting
1:03:16all individuals within a randomly
1:03:18selected collection
1:03:20of groups selected collection or
1:03:23group of individuals okay so the
1:03:26population
1:03:27is either broken up into little groups
1:03:30or maybe they are
1:03:31naturally clumped up in certain groups
1:03:35by certain characteristics
1:03:36and then you're going to randomly pick
1:03:38one of those groups okay
1:03:40so for example if i wanted to take a
1:03:43survey
1:03:44of students right if i wanted to do it
1:03:47simple random sampling
1:03:48i would go to the computer programs at
1:03:50school right and get a
1:03:52list of all 20 000 students and maybe i
1:03:55randomly select 20 of them i
1:03:58have the computer randomly select 20 of
1:04:00them and then i track them down
1:04:02i call them i email them i go to their
1:04:05house i go to their job i
1:04:07track those people down and go you were
1:04:09selected to be a member of this
1:04:11sample please let me get some
1:04:14information from you
1:04:15right that would be simple random
1:04:17sampling
1:04:18on the other hand if i wanted um
1:04:21a cluster sampling what i could do
1:04:24is um go to a particular classroom right
1:04:29i know right now it's covered era so we
1:04:31don't actually have classrooms but
1:04:33ignoring this little scenario for this
1:04:35little scenario let's pretend
1:04:36everything's normal
1:04:38so everyone is already in classrooms
1:04:41so it would just be super easy for me to
1:04:43randomly pick a classroom
1:04:45right randomly pick any room any
1:04:48classroom in the entire campus
1:04:50randomly walk into one of those
1:04:51classrooms and survey everybody
1:04:53in that classroom so that would be uh
1:04:56cluster sampling good
1:05:01so if the members of the population are
1:05:03already broken up into some sort of
1:05:04group
1:05:05and then you randomly pick a group and
1:05:07then that group is
1:05:08a member of your sample then then that
1:05:12would be
1:05:12an example of cluster sample
1:05:16so a school less matter wants to obtain
1:05:18a sample of students in order to conduct
1:05:20a survey
1:05:20she randomly selects 10 classes and
1:05:23administers the survey to all the
1:05:24students
1:05:25in those classrooms right so that's
1:05:26cluster sampling
1:05:30here's a nice little
1:05:36nice little uh visual representation of
1:05:40each of them
1:05:41right so in simple random sampling we
1:05:44have our population
1:05:46and any group of people can be selected
1:05:48to be members
1:05:49of my sample right in stratified
1:05:52sampling
1:05:53first you separate the population into
1:05:55strata
1:05:56that have some sort of homogeneous
1:05:58characteristics something similar about
1:06:00them like these are all the men these
1:06:01are all the women
1:06:03for example or it could be broken down
1:06:05by race
1:06:06or it could be broken down by um that by
1:06:09degree
1:06:10right what what what their uh stated um
1:06:15degree is but what degree they're trying
1:06:17to achieve
1:06:18um it could be broken down by highest
1:06:22level of
1:06:22math completed it could be broken down
1:06:25by
1:06:26um i don't know zip codes where they
1:06:29live uh could be broken down by lots of
1:06:31different ways so first you set up the
1:06:32little groups
1:06:34and then from each individual group you
1:06:36grab people
1:06:38that uh through through the process of
1:06:40simple random sampling within each
1:06:42strata
1:06:43and then that's how you get your final
1:06:45sample of people
1:06:46right thereby guaranteeing some sort of
1:06:48makeup like in this case we are
1:06:50guaranteeing that we want two males two
1:06:52females
1:06:53right or maybe we want something
1:06:55different maybe we want four women
1:06:57and two men right so this this process
1:07:00would guarantee that
1:07:02okay yeah question uh yeah
1:07:05so in chapter one we have a population
1:07:08a sample and an individual so
1:07:11here in chapter 1.4 we have the
1:07:14population sample
1:07:15or the strata with the strategy i see
1:07:18sample there what part of the strata
1:07:20would fall
1:07:21from chapter one or would it be a whole
1:07:23different subject
1:07:24from the population the sample and the
1:07:26individual
1:07:28okay so let me see if i can answer that
1:07:30so strata
1:07:31just means that you're
1:07:34you're separating your population so
1:07:37this is still part of the population
1:07:39you're just coming up with some way of
1:07:41separating them
1:07:42first before you pick your members and
1:07:45then
1:07:46this over here is your actual sample
1:07:49okay so your population
1:07:52um see here we we didn't break them up
1:07:54in any kind of way so we could just
1:07:56randomly pick anybody
1:07:57so you have no real control over what
1:08:00happens with your sample it could be
1:08:03lots of males lots of females it could
1:08:05be lots of
1:08:06business majors it could be lots of
1:08:08people that live in zip code
1:08:10you know zero zero two zero four you
1:08:13know you have no control it could be
1:08:14anybody from in there
1:08:16simple random sampling allows all
1:08:18possible groups of that size
1:08:20any three people could have been chosen
1:08:22no control
1:08:23so if you want to have some control like
1:08:25maybe
1:08:26this school happens to be situated
1:08:29in the middle of like three major zip
1:08:32codes
1:08:33this is zip code zero zero one this is
1:08:35zip code zero
1:08:36zero zero two and zero zero
1:08:39zero three and maybe this is like the
1:08:42rich neighborhood
1:08:43and this is the not so rich neighborhood
1:08:45and this is like the middle
1:08:47and your school happens to be like kind
1:08:49of like there
1:08:50here's my school um okay well
1:08:54you want a sample of students but maybe
1:08:56you want to make sure
1:08:57that they're all equally represented so
1:08:59you want three students
1:09:01from each zip code okay so you take your
1:09:04population of students
1:09:06you break them up into groups of zip
1:09:08codes based on zip code
1:09:10and then you group you take samples from
1:09:12each individual group
1:09:14thereby guaranteeing that your sample
1:09:16has a particular makeup
1:09:19okay yeah thank you
1:09:23okay all right so the strata is the
1:09:26thing you use
1:09:27to separate your population it could be
1:09:29zip codes it could be gender it could be
1:09:31race it could be your major
1:09:34it could be height it could be anything
1:09:37okay
1:09:37any characteristic that that can
1:09:39separate these people and then you take
1:09:41your sample
1:09:42systematic is when you're able to
1:09:45organize
1:09:46all the people in a line or maybe they
1:09:48are already like that so right so that
1:09:49classic idea
1:09:50is people single file coming out of an
1:09:53airplane
1:09:54or coming out of a movie theater they're
1:09:56already in like single file
1:09:58so it just makes it convenient to stop
1:10:01some people and ask them to be part of
1:10:03the survey
1:10:04good so there's a little process that
1:10:05you follow to figure out how to pick
1:10:07those people
1:10:08so what you're saying here is okay i'm
1:10:10gonna pick number two person number two
1:10:12that
1:10:12walks out of the movie theater or person
1:10:14number two that walks out of the
1:10:16airplane
1:10:16or person number two that walks out of
1:10:18the math building and starting at eight
1:10:20a.m
1:10:21right you camp out it's starting at 8
1:10:23a.m
1:10:24and you're gonna okay as people walk out
1:10:26of this building there's only one exit
1:10:28i'm gonna grab the second student and
1:10:30then my account and then the fifth
1:10:32student
1:10:32and then i'm going to count and then the
1:10:34eighth student and then the 11
1:10:36student and there's my sample
1:10:39in a sort of a systematic sampling of
1:10:42people
1:10:46um if you do it this way you're really
1:10:49not guaranteeing that you're going to
1:10:50get
1:10:51that any group of people uh any final
1:10:54sample
1:10:54is equally likely of occurring because
1:10:57for example people
1:10:59leave a um
1:11:02people leave a plane in a particular
1:11:05order
1:11:06right there's something there's some
1:11:07common characteristics of people
1:11:09so for example in most planes that still
1:11:13have
1:11:13um you know classes first class is in
1:11:17the front
1:11:17usually they get that privilege of
1:11:20leaving the airplane first
1:11:22right so if you're just standing out and
1:11:25watching people walk out of the airplane
1:11:27and you want to get the first person and
1:11:29then the fifth person
1:11:31and then the eighth person and then the
1:11:33eleventh person
1:11:34um there's a small there's there's
1:11:37there's zero
1:11:38chance that you're gonna get the first
1:11:40four people
1:11:41in your sample that is not
1:11:45one of the possible surveys that you're
1:11:47gonna get because you're doing
1:11:49systematic
1:11:50so therefore it's not true that all
1:11:53samples
1:11:54are possible and it's not true that all
1:11:56samples are equally likely of occurring
1:11:58the way they the way they would be if i
1:12:00did it this way
1:12:01if i took everybody's name in the
1:12:03airplane and i said i'm going to
1:12:04randomly pick three people
1:12:06it's possible that it could be that
1:12:08those three people are all in first
1:12:10class
1:12:10and they were all in seat number one two
1:12:12three maybe they were all related
1:12:14it's possible simple random sampling
1:12:16allows that possibility
1:12:18but with systematic uh sampling it
1:12:21severely reduces that right because it's
1:12:24more likely that first-class passengers
1:12:26all kind of come out first
1:12:27not guaranteed maybe one of them you
1:12:29know kind of sat behind and waited and
1:12:31then maybe they're way back here
1:12:33or maybe they like waiting for the whole
1:12:34plane to empty and they can go out and
1:12:36be comfortable and not be rushed or
1:12:38whatever
1:12:39so it's not guaranteed it's more likely
1:12:41that people in first class kind of walk
1:12:43out first
1:12:44and it's also more likely that related
1:12:46people kind of walk out together
1:12:48so it's more likely that these three
1:12:50people
1:12:51are somehow flying together um and
1:12:54that's why they
1:12:55they are leading together or sort of
1:12:57clumped up
1:12:58so anyway not all light uh samples are
1:13:00equally likely of occurring and not all
1:13:02samples are even possible
1:13:04when you do systematic sampling good
1:13:08so simple random sampling is the gold
1:13:10standard if you can do it
1:13:11but logistically there's several reasons
1:13:15for why it's not always
1:13:17the the best thing to implement but you
1:13:19do get the best results
1:13:21on the other hand sometimes you
1:13:23specifically want a particular makeup
1:13:25like i said if you want um some students
1:13:28to represent the school to be
1:13:30interviewed on on tv you maybe
1:13:33specifically want something you want an
1:13:35equal number of men and women or maybe
1:13:37you want
1:13:38uh particular um majors right we want
1:13:42we want some business we want some
1:13:43engineering right so you might
1:13:45want to interfere in your sample and
1:13:47make sure your sample has a particular
1:13:49makeup
1:13:50good and then finally cluster sampling
1:13:53sampling
1:13:54you take your whole population and
1:13:56instead of creating strata
1:13:57right here strata is because they have
1:13:59some characteristic like men versus
1:14:01women
1:14:02right but instead what you do is you
1:14:03take your whole population
1:14:05and you think about how they're broken
1:14:06up in some so into natural groups or
1:14:09clusters
1:14:10like classrooms right this is classroom
1:14:12one classroom two
1:14:14classroom three they're already broken
1:14:16up by certain groups
1:14:17or maybe houses right we're about to
1:14:19take the u.s census
1:14:21so one of the things they do is they
1:14:22randomly pick houses and they go knock
1:14:24on them and they go hey
1:14:25can can i ask how many people live here
1:14:26and can i ask some information about who
1:14:28works and who doesn't work and
1:14:30you know that kind of thing so they
1:14:32packed with that last year
1:14:34what's that i did that last year
1:14:38well ten years ago i actually did that
1:14:40oh okay well
1:14:41they're about to do it again so um
1:14:45cluster sampling means that the entire
1:14:47population
1:14:48is already broken up into little groups
1:14:51um like a classroom or a house or
1:14:54something
1:14:55um or if you're an airline
1:14:58uh if you're american airlines and you
1:15:00want to survey people
1:15:02uh people are already grouped up into
1:15:04flights
1:15:05so maybe you randomly select one flight
1:15:08and then you go and you survey everybody
1:15:10on that plane
1:15:12uh and so that's cluster sampling
1:15:15okay any questions about the differences
1:15:23there's a lot of controversy uh with the
1:15:26senses in particular
1:15:28uh because um
1:15:31originally well not originally but the
1:15:34intention
1:15:35some people feel that the intention has
1:15:37always been
1:15:38that they literally count people you
1:15:41know
1:15:41one two three four count them how many
1:15:44are there
1:15:45but that's hard to do uh especially as
1:15:48our population has grown bigger and
1:15:50bigger and bigger and bigger
1:15:51right we're expecting that our census
1:15:54this year will probably be a little
1:15:56higher than 320 million people
1:15:59so that's a lot of people to count so
1:16:02instead
1:16:03we have relied more and more and more on
1:16:06statistical um theory
1:16:09to help us get a sense get an idea
1:16:13of what that number is rather than
1:16:15actually count people
1:16:17um and that's been a source of
1:16:18controversy uh
1:16:21because obviously uh with anything we do
1:16:24in math there's always going to be
1:16:26some level of error right i mean we
1:16:29could be wrong we're
1:16:30we're gonna count literally count some
1:16:33of the people
1:16:34in uh in in a way that is consistent
1:16:37with
1:16:37statistical theory and then we're going
1:16:39to use that
1:16:41those results to make really
1:16:44really really good guesses about the
1:16:47entire population
1:16:48but no matter how good that guess is
1:16:50it's always going to be
1:16:51a guess so there will always be
1:16:55some level of error and some people
1:16:58aren't happy about that
1:16:59so the controversy is
1:17:02should we make an attempt to literally
1:17:06literally count people one two three
1:17:08four five count them
1:17:10or should we devote our resources
1:17:13to only counting some of them
1:17:16but in a way that is consistent with
1:17:19statistical theory
1:17:20so that we can use math and statistics
1:17:23to get a really good idea of how many
1:17:26people
1:17:27live in this country and where they live
1:17:29which state they live in
1:17:31right so that's something to think about
1:17:33you know is it
1:17:34is it worth it some people are very
1:17:36unhappy because obviously
1:17:37what if we make a mathematical mistake
1:17:40and then
1:17:41we could be wrong um but then again you
1:17:43know when you're actually literally
1:17:44counting people
1:17:46there's a lot of room for mistakes there
1:17:48too so
1:17:50you know i don't know something
1:17:52something to think about
1:17:53right when you're literally counting
1:17:54people it's really hard to do that
1:17:56um especially you know there's a lot of
1:17:59groups of people that are really hard to
1:18:00count
1:18:01you know particularly the homeless for
1:18:02example uh
1:18:04people that people that work a lot
1:18:06they're they're hardly ever home
1:18:08um and there's a lot of people that kind
1:18:10of live off the grid they
1:18:11live pretty far away from everyone else
1:18:14um so it's kind of hard to find them
1:18:16um you know it's it some people work
1:18:19pretty hard
1:18:20to not be uh in the system right not be
1:18:24on the grid
1:18:25uh so it's kind of hard to know where
1:18:27they are and how many there are and all
1:18:29that
1:18:30right there's a lot of people that don't
1:18:32trust
1:18:33when someone comes knocking at the door
1:18:35and asks questions about who lives there
1:18:37and whatnot they're not going to answer
1:18:38the door they're not going to answer
1:18:39your questions they don't want to
1:18:40participate they don't trust you
1:18:42they don't know what this is about and
1:18:45so it makes it really hard to literally
1:18:47count people
1:18:48so who knows the debate goes on
1:18:54good good good okay other effective
1:18:57sampling methods okay stratified and
1:18:59cluster samples are different
1:19:00uh in a stratified sample would divide
1:19:02the population into two
1:19:04or more homogeneous groups right
1:19:05homogeneous means that they have
1:19:07some particular characteristic in common
1:19:09so this is the
1:19:10males versus females for example then we
1:19:13obtain a sample
1:19:14a simple sample a simple random sample
1:19:17from each group
1:19:18in a class cluster sample we divide the
1:19:20population into groups obtain a
1:19:22simple random sample of some groups
1:19:26right so
1:19:26and survey all the individuals in the
1:19:28selected groups so we
1:19:30randomly pick a group and then survey
1:19:32everybody in that group
1:19:37so again with the airline for example if
1:19:40i'm
1:19:40um on the board of uh
1:19:44you know on the board of american
1:19:46airlines and i want to find out about my
1:19:49customers
1:19:50uh one thing i could do is maybe i want
1:19:52to survey
1:19:55i want a survey that that that tells me
1:19:57information about my customers
1:19:59but maybe i don't just want any random
1:20:01sample of anybody
1:20:02that has ever flown on my airline i want
1:20:05to
1:20:06maybe focus on where the money is right
1:20:09i mean
1:20:09my my frequent flyers those are the
1:20:12people that i care about the most
1:20:14so maybe i want to separate
1:20:17my list of contact information right
1:20:19because whenever you
1:20:20buy a ticket you give them your email
1:20:22and stuff so i have that all that
1:20:24contact information
1:20:25but maybe i want to separate it so that
1:20:28i have
1:20:29people that are frequent liars people
1:20:31that that fly
1:20:33at least once a week there's people that
1:20:35fly once a week
1:20:36you know for months um so people that
1:20:39fly once a week
1:20:40maybe that compare it to people that
1:20:42don't fly that often but maybe spend a
1:20:44lot of money
1:20:45so maybe they're the first class people
1:20:47that that dropped the big bucks on the
1:20:49big flights um one time i flew to
1:20:53from lax to um
1:20:58i forgot what that i think was like hong
1:21:00kong maybe anyway
1:21:02it wasn't my final destination but it
1:21:04was one of the flights
1:21:06on you know kind of moving on to where
1:21:08i'm going but anyway the big leg of the
1:21:10flight was from la to hong kong
1:21:12um and i think my ticket was like twelve
1:21:16hundred dollars or something like that
1:21:18for coach kind of ticket uh
1:21:21and then shortly after i got an email
1:21:23that invited me
1:21:25to upgrade to a first class ticket um
1:21:28for um a nominal fee upgrade right i was
1:21:32like oh okay a nominal fee upgrade
1:21:34should i consider that my ticket was
1:21:36like twelve hundred dollars
1:21:37how much more could they possibly want
1:21:39to upgrade to first class
1:21:42it was like ten thousand dollars they
1:21:44emailed me if i wanted to
1:21:46spend an additional ten thousand dollars
1:21:49to upgrade the first class on that
1:21:51flight
1:21:53airplanes are ridiculous yeah so i was
1:21:56like
1:21:56really 10 no no thank you but i mean it
1:21:59occurred to me like somebody out there
1:22:00does that
1:22:00like somebody's spending 10 grand
1:22:04on on a flight to be in first class
1:22:07so anyway um the airline might really
1:22:10care about hearing from those people
1:22:13a lot more than they want to hear from
1:22:14someone like me that just buys whatever
1:22:16the cheapest ticket is
1:22:17um so uh in order to
1:22:21to hear from their customers they might
1:22:24divide up their population right the
1:22:26population being
1:22:27all uh all previous
1:22:31passengers for whom they have contact
1:22:33information
1:22:34um and so they might want to divide them
1:22:36into little strands
1:22:38first class people business people
1:22:40frequent flyer people
1:22:42and then everyone else you know coach
1:22:44people that don't fly that often
1:22:47okay and then maybe they then they want
1:22:49to uh
1:22:50ran simple random sample people from
1:22:53each group
1:22:54right that's stratified sampler on the
1:22:57other hand
1:22:57cluster sampling they might just go into
1:23:00their computer system
1:23:01and they randomly pick three flights
1:23:05anywhere in the world right at any
1:23:06moment they have thousands of flights
1:23:08all over the world so maybe they just
1:23:11randomly pick three flights that are
1:23:12you know gonna happen today and then
1:23:15they go to those airplanes
1:23:17and you know before they take off you
1:23:19know okay we're about to take off
1:23:21or maybe in midair that's a good place
1:23:22to do it you've got them trapped in
1:23:24midair
1:23:25you know midair you go on the mic and go
1:23:27okay
1:23:28you've been all uh selected to be part
1:23:32of this
1:23:32survey please fill out these surveys
1:23:36and uh i don't know if you as a reward
1:23:38will give you
1:23:39whatever free peanuts or something an
1:23:41extra an extra soda
1:23:43something uh but anyway now you're
1:23:45getting information from everybody
1:23:47in one airplane rather than this other
1:23:50method
1:23:51of breaking people down and having some
1:23:53control
1:23:55okay any questions about the differences
1:24:00no no another caution about convenience
1:24:04sampling
1:24:05right if the members of the population
1:24:08self-select
1:24:09to be members of the sample um
1:24:12that's not good right a convenience
1:24:14sample is one in which the individuals
1:24:15in the sample are
1:24:16easily obtained as in they volunteer
1:24:19that's one way to do it
1:24:20or you just grab the first few
1:24:22convenient low-hanging fruit kind of
1:24:24thing
1:24:24um i wanted to get a survey
1:24:28of of people so how about if i just
1:24:30survey the people that are here right
1:24:32now
1:24:32you guys are here congratulations you're
1:24:34part of my sample
1:24:36right that doesn't that's not convenient
1:24:38i'm sorry that's very convenient
1:24:39so that's not going to do a good job of
1:24:43actually representing the population
1:24:47right so uh uh any studies that use this
1:24:49type of sampling uh generally results
1:24:52are suspect uh results should be looked
1:24:54upon with
1:24:55extreme skepticism okay multiple
1:24:59uh multi-stage sampling right so we can
1:25:02kind of combine
1:25:03these things uh and and come up with
1:25:06different
1:25:06um uh more complicated ways of getting
1:25:09your sample where you have a combination
1:25:11of different different
1:25:12types so in practice most large-scale
1:25:15uh surveys obtain samples using a
1:25:17combination of techniques just presented
1:25:19as an example of multi-stage sampling
1:25:22consider the
1:25:22the nielsen media research right so uh
1:25:26nielsen randomly selects households and
1:25:29monitors the television programs these
1:25:31households are watching
1:25:32through a people meter the meter is an
1:25:35electronic box
1:25:36uh placed on each tv within the
1:25:39household the
1:25:40the people meter measures what program
1:25:42is being watched and who is watching it
1:25:44okay well that's kind of weird so it's
1:25:46got like a camera maybe pointing at the
1:25:48couch
1:25:49and it's like keeping track of how many
1:25:51people are watching and who is watching
1:25:53it
1:25:54um so i'm sure these people um
1:25:57self-select to be part of this right
1:25:59because you can't randomly
1:26:01um well i guess maybe you can you can
1:26:04ask them
1:26:05and then maybe they choose to be
1:26:06participants in this
1:26:10uh nielsen selects the households uh
1:26:13with with use of a two-state sampling
1:26:15process so stage one
1:26:16using u.s census data this divides the
1:26:19country into geographic areas
1:26:21strata the stratas are typically city
1:26:23blocks in urban areas
1:26:25and geographic regions in rural areas
1:26:28about 6 000 strata are randomly selected
1:26:31so they
1:26:31subdivide the whole country into little
1:26:33groups
1:26:35based on whether you live in an urban
1:26:37area whether you live
1:26:38in a uh you know city certain city
1:26:42blocks
1:26:43uh nielsen then sends representatives to
1:26:45the selected strata
1:26:46and lists the households within the
1:26:48strata the households are then randomly
1:26:51selected through a simple random sample
1:26:55nelson sends um this and sells the
1:26:58information
1:26:58obtained to television stations and
1:27:00companies this was also used to help
1:27:02determine prices for commercials
1:27:04right they need to figure out who's
1:27:05watching it and
1:27:07um and then they use that information
1:27:11to be able to uh assess the value
1:27:14of a particular tv show right if a tv
1:27:18show attracts a lot of people uh
1:27:21then then that tv station can command
1:27:24more money for their commercials
1:27:26and not just how many people but also
1:27:30they like to get information about the
1:27:32ages of the people
1:27:33watching it um and males versus females
1:27:37so that commercials can be more targeted
1:27:40right maybe it's a product that sells
1:27:42particularly well
1:27:44to a particular group right so like
1:27:46video games for example
1:27:48traditionally sell best to
1:27:51uh people in the age ranges of like
1:27:5418 to 35 in that sort of age range
1:27:58um you might wonder what about below 18
1:28:01they play a lot of big
1:28:02games true but they don't have any money
1:28:04uh so we want people that play a lot of
1:28:06video games and have the money to buy
1:28:08them
1:28:08uh so that that could be your target
1:28:10right and
1:28:11above 35 they finally get a life so they
1:28:14don't play as many video games
1:28:16hopefully so maybe that's your target so
1:28:20you're looking for a tv show that
1:28:21attracts a lot of people
1:28:23in that age range and also studies show
1:28:26it's mostly men that
1:28:28that uh play video games and um and buy
1:28:31the video games so you want males 18 to
1:28:3335
1:28:34so you want a tv show that attracts
1:28:36people like that i don't know
1:28:38maybe american ninja warrior might be a
1:28:41tv show that does really well with that
1:28:43age group
1:28:45good questions questions questions
1:28:50okay in this last section we want to
1:28:51talk a little bit about bias
1:28:53biases um so
1:28:57some definitions so if the results of
1:29:00the sample are not represented of the
1:29:02population
1:29:03then the sample has bias right so
1:29:06if we do a good job
1:29:09of selecting our sample through a random
1:29:12process
1:29:13and our sample does a really good job of
1:29:15representing the population
1:29:17in all aspects like for example
1:29:21if you think about our school all twenty
1:29:23000 students and i randomly pick
1:29:25100 that 100 if that random sample of
1:29:28100 was all
1:29:29male that would obviously not be a good
1:29:33representation of the entire school
1:29:35right or i randomly pick 100 students
1:29:38and all 100
1:29:40um are business majors that would not do
1:29:43a good job of representing the entire
1:29:45school
1:29:46right so we want a little microcosm a
1:29:48little micro universe
1:29:50that represents the the whole school
1:29:54so it should do a good job of having
1:29:56about the same makeup
1:29:58of males to females about the same
1:30:00makeup of
1:30:01students that have children versus that
1:30:04are parents
1:30:04uh students that are parents versus
1:30:06non-parents uh a pretty good job of
1:30:08representing
1:30:09uh a percentage of students that have a
1:30:12full-time job versus not a full-time job
1:30:14of students that commute
1:30:16long-distance students that you know
1:30:18just just about everything right
1:30:20it should do a good job of representing
1:30:22the school
1:30:23if it fails to do that if there's some
1:30:25characteristic
1:30:26that is very very different in our
1:30:28sample of students
1:30:30versus the entire school then there's a
1:30:32bias that that exists there
1:30:35okay um so that bias could be there
1:30:39on purpose somebody intentionally
1:30:41created that bias
1:30:43or it could be there accidentally but
1:30:46whether it's there on purpose or
1:30:47accidentally
1:30:48the fact that it's there uh you know
1:30:51nonetheless you have bias
1:30:52so um when you have bias
1:30:56then then your results aren't going to
1:30:58be as good right
1:30:59your your sample doesn't do a good job
1:31:01of representing the entire population
1:31:04so it the results uh aren't going to be
1:31:07very good
1:31:09and when i say the results i should make
1:31:10that clear right our goal in statistics
1:31:13is to have you know we have a giant
1:31:16population
1:31:18and we want to know something about this
1:31:20population um
1:31:22we want to know if they're going to vote
1:31:23yes or no
1:31:25on a certain thing right it doesn't
1:31:28matter thing
1:31:30a how do they feel about thing a yes or
1:31:33no
1:31:34on it i don't know right and there's a
1:31:36lot of people here there's 20
1:31:37000 people so it's hard to really know
1:31:40how everybody feels about this thing yes
1:31:42or no
1:31:43so what we do is we take a little sample
1:31:45that's more manageable
1:31:47and we're going to take that sample
1:31:48maybe that sample only has 100
1:31:50right so this is capital n i should say
1:31:53that
1:31:54capital n is 20 000. the number of
1:31:58members of my population
1:32:00and in my sample we use lowercase n
1:32:04maybe my
1:32:04sample only has 100
1:32:10right and now it's a lot more manageable
1:32:12to ask about thing a here
1:32:14yes or no what do you guys think thing a
1:32:16yes or no
1:32:17okay so what i'm gonna do is i'm gonna
1:32:19take information from here
1:32:21okay so from here maybe i get that uh
1:32:2558 say yes
1:32:28and um and 40 say no
1:32:34and that means that two they have no
1:32:38idea
1:32:38i don't know what i'm talking about what
1:32:40thing there's a thing a never heard of
1:32:41it
1:32:42okay um so this is the results we get
1:32:45there's no controversy here this is
1:32:48exactly the results right
1:32:50no doubts at all this is what i got
1:32:53but the point is that we want to use
1:32:55this information from my sample
1:32:57to make a good guess about these people
1:33:02right the 20 000 based on the 20 000
1:33:06it looks like okay maybe about
1:33:0958 of these people are gonna say yes
1:33:13maybe um or or can i at least say
1:33:16more than half are going to say yes
1:33:19maybe
1:33:20right so i want to do that but i want to
1:33:22it's definitely going to be a guess but
1:33:23i want that guess to be
1:33:25as good as it can be as accurate as it
1:33:28can be
1:33:29um and that's the whole that's the most
1:33:32important part of statistics that's what
1:33:33we're going to be doing
1:33:34making guesses about our population
1:33:37based on information from our sample
1:33:39but in order for this guess to be good
1:33:42the first thing we have to make sure is
1:33:43that it doesn't have any bias
1:33:45we want to make sure that this sample
1:33:47does a really really good job
1:33:49of representing this population if it
1:33:51doesn't do that
1:33:52if it's missing a group of people or if
1:33:55it's over
1:33:56representing a group of people then when
1:33:59we get these results
1:34:00they may not do a good job of accurately
1:34:04representing these people right it may
1:34:06not give you a good guess
1:34:09does that make sense so if i got my
1:34:12sample
1:34:13by say back to the students i got my 100
1:34:17uh student sample by going on campus
1:34:20and randomly picking a hundred students
1:34:22that are walking around
1:34:24okay but that's not going to capture the
1:34:27students that take classes online
1:34:29they're not walking around then right so
1:34:32that's not going to give you a good
1:34:33sample
1:34:34so when i get information about things
1:34:36that's not necessarily going to give you
1:34:38a good guess about how the entire school
1:34:40feels because your sample
1:34:42didn't have any students that only learn
1:34:44online
1:34:45good or on the other hand um you know
1:34:48again i go to
1:34:49i go to school and i randomly pick
1:34:51people
1:34:52but i do it in the morning then my
1:34:55sample is not going to include people
1:34:57that only take classes at night the
1:35:00night learners
1:35:02and you know night learners are probably
1:35:03more likely to have a full-time job
1:35:05that's why they're taking classes at
1:35:06night
1:35:07so if my sample only has people that
1:35:09take classes in the morning and in the
1:35:11daytime
1:35:12then again the results from that sample
1:35:14are going to do a poor job
1:35:16of generating guesses for what the
1:35:19entire population feels
1:35:23good any questions
1:35:26no no okay so we're going to
1:35:29look at or sub categorize biases into
1:35:33three general things there's sampling
1:35:35bias
1:35:36non-response bias and response bias
1:35:41okay sampling bias means that the
1:35:43technique used to obtain the individuals
1:35:45to be in the sample tends to favor one
1:35:48one part of the population over another
1:35:50right so this is like i was saying
1:35:52i just go to campus at eight in the
1:35:54morning and i sample
1:35:56100 students that are walking around
1:35:58okay
1:35:59well i'm not taking into consideration
1:36:02students to take night classes
1:36:03so my sample isn't going to be
1:36:05reflective of the entire school
1:36:07i'm not taking into consideration
1:36:09students that only take classes online
1:36:11so my sample is not going to include
1:36:13those people got it so
1:36:16the way in which i selected members of
1:36:19my sample
1:36:20is intentionally favoring one group of
1:36:23people
1:36:24or or excluding a group of people
1:36:28good um
1:36:31maybe uh in order to
1:36:35participate in my sample
1:36:38what can i do what could i do maybe i
1:36:41decide
1:36:41to to um put flyers on the windshields
1:36:45of people
1:36:46that could be one thing i do uh you know
1:36:49people are busy they're walking around
1:36:50they don't want to let me stop them and
1:36:52talk to them they're on their way to
1:36:53class
1:36:54so maybe what i do is i go to the
1:36:55parking lot and i put a little
1:36:57sample thing on the windshield of cars
1:37:00um and even if i keep track of them
1:37:03maybe i keep track of their license
1:37:04plate and i go to the school computer
1:37:06and i figure out
1:37:07license plates and contact info and then
1:37:09i email them and go hey i left the
1:37:11survey
1:37:11on your windshield can you please
1:37:13participate or whatever
1:37:15i mean even if you keep track of them
1:37:16like that you're
1:37:18creating a bias right because you're
1:37:21intentionally
1:37:22only including people in your sample
1:37:25that drove
1:37:26and parked their a car to school
1:37:29what about all the people that walked on
1:37:30campus or the people that live on or
1:37:32near campus
1:37:34aren't going to be representative in
1:37:35your sample uh all the people that took
1:37:37public transportation are not going to
1:37:39be in your sample
1:37:40good so some sort of bias is there your
1:37:43sample
1:37:44intentionally favors one group of
1:37:46students over another
1:37:48um right or maybe
1:37:51in order to be a member of my sample
1:37:55i ask everybody to come to the theater
1:37:59you know come to the theater at 8 pm on
1:38:01a monday
1:38:02and you know we're gonna get everybody
1:38:04together and we're gonna ask people
1:38:06questions and
1:38:07measure things and you're gonna be part
1:38:09of our sample
1:38:10you know please come join us well again
1:38:12you're creating a bias
1:38:13because for some people it might be
1:38:15really difficult to go to campus
1:38:17at eight o'clock at night and for some
1:38:19people it's really easy if you live
1:38:21in on campus or near campus then it's
1:38:24not that big of a thing to walk on
1:38:26campus again
1:38:27and uh go to this big meeting um
1:38:30but if you live really far away if you
1:38:33have to commute
1:38:34traffic work like all kinds of other
1:38:37things
1:38:38could limit your participation in that
1:38:40and thereby
1:38:41favoring one group of people over
1:38:43another
1:38:45good there's a really famous example
1:38:48uh i think it's uh dewey i think was his
1:38:52name
1:38:52uh uh in the 19th
1:38:5852
1:39:00is that 1952 no that couldn't have been
1:39:0352
1:39:0548 i think it was 1948 a presidential
1:39:09campaign
1:39:10um between dewey and truman right you
1:39:12can look that up
1:39:14uh they they did a really really
1:39:16extensive survey
1:39:17uh to to try and determine who was gonna
1:39:19win
1:39:20the presidency um and their
1:39:23methods uh led them to believe that
1:39:26dewey was going to win
1:39:29but of course truman won and they were
1:39:32so
1:39:32certain that dewey was going to win that
1:39:35that
1:39:36newspapers ran with a
1:39:39headlines that you know breaking news
1:39:42dewey won the presidency before the
1:39:46the final vote was in uh so anyway there
1:39:49was just a lot of biasy
1:39:51uh biases in their um
1:39:54in their technique i think what happened
1:39:56with them is that they
1:39:58they selected people to be a member of
1:40:01their sample
1:40:02by going through dmv records so they
1:40:05they went to the dmv they got um
1:40:08addresses of people and then they mailed
1:40:10them
1:40:10a little survey who do you think is
1:40:12gonna win who are you gonna vote for
1:40:13and they did it that way um and so there
1:40:16was a lot of biases
1:40:17especially uh in you know 1948
1:40:20in order to have a car in order to have
1:40:22a driver's license you kind of had to be
1:40:24a little bit more well off right
1:40:27cars were very very expensive relatively
1:40:30speaking
1:40:31a lot of people couldn't afford a car
1:40:33they couldn't afford to have
1:40:34a driver's license so it there was a
1:40:38bias there
1:40:39the process by which they chose uh
1:40:41people in their sample
1:40:43favored um people that were a little bit
1:40:45more well off
1:40:46financially and it didn't it didn't
1:40:49include a whole
1:40:50lot of other people so they got it wrong
1:40:52their sample
1:40:53did not do a good job of guessing what
1:40:55the entire population was thinking
1:40:58good good undercover results uh
1:41:01in sampling bias under coverage occurs
1:41:04when the proportion of one segment of
1:41:05the population
1:41:06is lower in a sample than it is in the
1:41:08population
1:41:09right so if
1:41:12your sample has a very small percentage
1:41:16of business majors and then you look
1:41:18around the population like wait a minute
1:41:20the most common major here is business
1:41:22how come our sample doesn't have that
1:41:23many business majors
1:41:24that would be under coverage right or
1:41:28uh you know say mexican maybe
1:41:31your campus has a lot of
1:41:36students that are mexican and then you
1:41:37take a sample and then it has a very
1:41:39small
1:41:40percentage of proportion of mexican
1:41:42students in your sample
1:41:43then that would be undercoverage right
1:41:46so a bias
1:41:46exists because your sample is not doing
1:41:49a good job of representing the entire
1:41:51population
1:41:53non-respond bias exists when individuals
1:41:55selected to be in the sample
1:41:57who do not respond to the survey have
1:41:59different opinions from those
1:42:00who do so if we do a good job
1:42:03of selecting simple random sampling i
1:42:06selected 100 students to be in my survey
1:42:09right but just because i selected them
1:42:11doesn't mean that they want to be in it
1:42:13right like you get an email that says
1:42:15congratulations
1:42:16you have been selected to be part of
1:42:18this panel of students
1:42:20you might not want to participate
1:42:24so maybe you ignore all the emails and
1:42:26you don't
1:42:27you know you don't answer the call or
1:42:29maybe you say you do
1:42:30but then when it comes to actually
1:42:33filling out the survey
1:42:34and turning it in you don't you know
1:42:36that kind of thing so you just are
1:42:38non-responsive even though you've been
1:42:40selected they did a good job they the
1:42:42the the uh
1:42:44the people conducting the the survey did
1:42:47a good job of selecting you
1:42:48uh through simple random uh sample
1:42:51processes but
1:42:52you might not respond and it might be
1:42:56that people who don't respond uh tend to
1:42:59have an opinion that is a little bit
1:43:01different
1:43:01than everyone else that does and so
1:43:04therefore
1:43:04a bias again begins to to uh
1:43:09to to be created um
1:43:13for example um i know some people
1:43:18just don't choose to be so active
1:43:21and they happen to have you know some
1:43:23sort of characteristic
1:43:25um nothing good comes to mind but
1:43:29anyway i think that that's probably a
1:43:30pretty pretty
1:43:32um self-evident one
1:43:38okay uh response bias exists when
1:43:41the answers on a survey do not reflect
1:43:43the true feelings of the respondent
1:43:45right so for example um there's an
1:43:49interviewer error right so an
1:43:51interviewer
1:43:52is is um interacting with somebody
1:43:56that's a member of their sample they're
1:43:58asking them questions
1:43:59and maybe there's an error that happens
1:44:01there in communication
1:44:03right in the dialogue um there's there's
1:44:06a mist
1:44:06a misunderstanding of some kind and so
1:44:08therefore
1:44:10uh the response recorded
1:44:13doesn't truly reflect the feelings of
1:44:15the respondent
1:44:18misrepresented answers um
1:44:22so again maybe a mistake in which you
1:44:24know
1:44:25in in the way in which you express your
1:44:28feelings um you know a
1:44:31a lot of sometimes it's like you don't
1:44:33quite understand the options
1:44:35you're reading and you're like okay
1:44:36choose one of these four options and you
1:44:38don't quite understand it they're worded
1:44:40kind of weird like wait a minute um
1:44:44i don't understand which of these four
1:44:46options truly reflects my feelings
1:44:49uh so it could be a misrepresentation
1:44:52there
1:44:53um the wording on questions the actual
1:44:56phrase
1:44:56used um has an impact um
1:45:00we have found that
1:45:03the actual language used when you ask a
1:45:07question
1:45:08um impacts how people respond to it
1:45:11um so for
1:45:14example um
1:45:16[Music]
1:45:17if you if you intentionally
1:45:20phrase it um in a way that favors
1:45:25something
1:45:26um like if i said do you favor
1:45:30punishing polluters that hurt our
1:45:32environment
1:45:33or do you prefer to let them get away
1:45:35with it if i say it like that
1:45:38it it automatically is putting an
1:45:41emphasis
1:45:41on yeah we should punish them right um
1:45:44it would it would be
1:45:47difficult for someone to disagree with
1:45:49me to say yup let him get away with it
1:45:52um just the phrasing that i've used uh
1:45:55implies that i want a certain response
1:45:58from people
1:45:59and i'm likely going to get that
1:46:01response more often than not
1:46:03but the order of questions
1:46:06is also um important
1:46:10we know that you know you know it's a
1:46:12long list of questions so if you ask
1:46:14them about
1:46:15the environment and then you ask them
1:46:16about laws and then you ask them about
1:46:18fines
1:46:19in in the order in which you ask things
1:46:23uh might make a big difference so
1:46:27um if you ask a question about
1:46:31tree heights in the neighborhood right
1:46:32going back to that question
1:46:34should we allow anybody to
1:46:37plant any tree of any height and you
1:46:40know
1:46:41everyone should have the freedom to do
1:46:43that yes or no okay
1:46:44some people say yes some people say no
1:46:46another question could be
1:46:48when uh a person violates
1:46:52a law should they be fined
1:46:56uh you know should the city be allowed
1:46:57to find you up to
1:46:59ten thousand dollars for breaking one of
1:47:01the city ordinances
1:47:03now you're connecting money to it right
1:47:05so if
1:47:06that question um comes up first
1:47:10now you're thinking about money you know
1:47:12violating city ordinances and being
1:47:14fined a lot of money
1:47:15and now i ask you about this tree thing
1:47:17now you might think twice like hmm
1:47:20maybe we shouldn't be maybe we should i
1:47:22don't know
1:47:23it might impact your results so in one
1:47:26way or another
1:47:27uh so the order in which questions are
1:47:29asked might
1:47:30make a difference right so uh these are
1:47:33different ways in which
1:47:34uh a response bias might occur
1:47:42data entry error is an example so
1:47:45although not technically a result of
1:47:47response based data entry error will
1:47:49lead to results
1:47:50that are not representative of the
1:47:51population once uh
1:47:53data are collected the results may need
1:47:56to be entered into a computer which
1:47:57could result
1:47:58uh it could result in input errors right
1:48:00so the way
1:48:01to transform the data onto a computer
1:48:05one very famous example that we
1:48:08should consider since we have a
1:48:10presidential election coming
1:48:11coming around is that sometimes
1:48:16there is misunderstanding
1:48:19uh as to what the process is what the
1:48:22procedures
1:48:23are for submitting your your your
1:48:26uh your your ballot um so like there are
1:48:30some that have these little punch things
1:48:31you like punch a little hole
1:48:33where you want but if you turn the page
1:48:36the hole you just created causes
1:48:39problems right so you're on page one and
1:48:41you punch a hole
1:48:42you're feeling pretty good about it i
1:48:43punched there i punched there i punch
1:48:45there
1:48:45i feel good and now you turn the page
1:48:48but now the holes you just created
1:48:51make things look a little weird um so
1:48:54now
1:48:55there could be some confusion that's
1:48:57caused
1:48:58so that's that's one example of of a
1:49:02problem that could occur
1:49:04and then people don't follow the
1:49:06procedures so for example
1:49:08um in a particular voting ballot thing
1:49:12might require you to sign the bottom of
1:49:14each page
1:49:16page one i vote for this person this
1:49:18person this person
1:49:19sign at the bottom next page i vote for
1:49:22this person this person this person sign
1:49:24at the bottom
1:49:25next page there might be five pages and
1:49:27when you're done
1:49:28you got to close the book sign the front
1:49:31of the book
1:49:32put it in an envelope close the envelope
1:49:34sign the outside of the envelope
1:49:36and then put your your full address
1:49:38right that could be the process
1:49:40so what do we do if someone doesn't
1:49:43really follow
1:49:44every step like maybe somebody
1:49:47didn't sign the outside of the envelope
1:49:49but they signed every page on the inside
1:49:51what should we do should we count that
1:49:53ballot or should we go no didn't follow
1:49:55the rules let's throw it in the trash
1:49:58right or maybe they signed everything
1:50:00but they forgot to put their address
1:50:02uh in the front right or maybe they
1:50:06put something down they squiggled
1:50:07something but you can't make heads or
1:50:09tails out of it it just looks like a
1:50:11weird
1:50:11rambling you know it says put your
1:50:14address here and all you see is
1:50:16kind of like that los angeles county
1:50:18like
1:50:19i don't know is that even an address i
1:50:20can't tell so what do you do do you
1:50:22throw that ballot out or do you count it
1:50:25right
1:50:25and then you're supposed to sign each
1:50:27and every single page but what if you
1:50:28skip one page
1:50:30do you not count that particular page
1:50:33because you didn't sign at the bottom
1:50:35or do you throw the whole thing on the
1:50:36trash because the whole thing is invalid
1:50:38right so there's a lot of questions um
1:50:41you know
1:50:42to to be you know for people to think
1:50:44think through as you create your
1:50:46your ballots uh and a lot of
1:50:48opportunities for buyers
1:50:49biases to exist right where your your
1:50:53sample no longer does a good job of
1:50:55representing the population
1:51:02so non-sampling errors are errors that
1:51:04result from sampling
1:51:06biases non-response biases response
1:51:09biases or data entries
1:51:12or data entry errors such errors could
1:51:15also represent
1:51:16could also be present in a complete
1:51:19sense of the population
1:51:21so complete senses is when you ask a
1:51:24survey of
1:51:24every member of our population
1:51:32and versus a survey a survey is when you
1:51:35pick a group of people
1:51:36and then you ask them the members of
1:51:39that
1:51:40sample questions or get data from them
1:51:46sampling error is an error that results
1:51:48from using a sample
1:51:50to estimate the population estimate
1:51:52information about a population
1:51:54this type of error occurs because a
1:51:56sample gives incomplete information
1:51:58about a population okay so the sample
1:52:01doesn't do a good job of representing
1:52:03accurately representing
1:52:04the whole population

About this transcript

This page contains the full transcript of Day 2, Statistics, Ch1 sec2 to Ch1 sec5 by Eduardo Barajas, generated from the public captions YouTube serves with the video. The transcript has 16,991 words across 3,062 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to translate the transcript into 100+ languages.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, translation to 100+ languages, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.