Case Interview Fundamentals: Lesson 29 [Hypothesis Testing: the 3-step approach] — Transcript
Full transcript
- 0:00Hey guys, how are you doing today? In
- 0:02this video, I want to show you our
- 0:04three-step approach to hypothesis
- 0:06testing, the last building block in this
- 0:10course.
- 0:11Uh this is by far the hardest one to
- 0:13teach, but it's not necessarily the
- 0:15hardest one to do.
- 0:17And that's because it's where you can
- 0:18get the most
- 0:20creative. You can
- 0:22There are many, many possible answers
- 0:24that work, uh not only one or a few that
- 0:27are correct.
- 0:28Because you can do anything as long as
- 0:32you're actually testing the hypothesis
- 0:33in a feasible way.
- 0:35Our three-step approach it will help you
- 0:37to do that, but if you don't follow it,
- 0:39it doesn't mean you're going to get it
- 0:41wrong.
- 0:42So,
- 0:43uh
- 0:45because of this,
- 0:47the best way to learn this is to
- 0:48practice. This three-step approach is
- 0:50going to help you, but you're only going
- 0:54to learn by practicing a lot. And that's
- 0:56why you're have you're going to see
- 0:58several drills in this chapter, and
- 1:00that's also why I'm going to teach you
- 1:02by the end of this video how to come up
- 1:04with your own hypothesis testing
- 1:06follow-up questions to practice, okay?
- 1:11So, let's move on.
- 1:13The three-step approach involves
- 1:16uh step number one, defining your
- 1:18hypothesis, step number two,
- 1:21breaking it down, and step number three,
- 1:24gathering and processing data,
- 1:26uh and showing it the way you'd present
- 1:29it.
- 1:30Uh you're not necessarily going to do
- 1:33all the steps here.
- 1:34You
- 1:36might not need to define your
- 1:38hypothesis, or you might choose not to
- 1:40define your hypothesis to the
- 1:41interviewer. You may not need to break
- 1:44the hypothesis down into simpler
- 1:46hypotheses because maybe you're able to
- 1:49test that hypothesis straight away
- 1:52instead of breaking it down into simpler
- 1:53hypotheses.
- 1:54Maybe you won't need to show the
- 1:57interviewer how you would gather and
- 1:58process data
- 2:00or at least you just want to show them
- 2:03some crucial things of how you do that
- 2:05and not others.
- 2:07As long as you tell them what you need
- 2:10to be closer to proving or disproving
- 2:11the hypothesis and how you'd get it, you
- 2:13are fine. So, this is different from
- 2:15estimations, for example.
- 2:17In an estimation, if you don't do your
- 2:20structuring or if you don't do the
- 2:22reality check, the interviewer is
- 2:24actually going to take points away from
- 2:26you for that. In the hypothesis testing
- 2:29questions, as long as you get what you
- 2:33tell them how you're going to get what
- 2:34you need and what you need,
- 2:37um you're fine. You're do you're going
- 2:39to do well in this question. Okay?
- 2:42Now, I want to show you
- 2:44uh a case question to which you can come
- 2:47up with your own answer for.
- 2:49And then I want to show you the
- 2:52follow-up questions, which is the
- 2:53hypothesis testing question. And as I
- 2:55said, as I've done in other
- 2:58uh
- 2:59uh step-by-step videos, I'm not going to
- 3:01show you the answer for the framework
- 3:03question, only for the follow-up
- 3:04questions. But, feel free to come up
- 3:07with your with your answer for it um cuz
- 3:11it might help you with your hypothesis
- 3:13testing. You might You might have
- 3:15actually covered the hypothesis question
- 3:17test hypothesis testing question in it.
- 3:19Um and you're still getting more
- 3:22practice for framework questions. And
- 3:24this is a cool framework question. So,
- 3:27DSC is a large drugstore chain in Brazil
- 3:30with 10% market share,
- 3:321,100 stores,
- 3:35and 9 billion Brazilian reais in
- 3:37revenues. The company faces challenges
- 3:40with losses for as long as they can keep
- 3:42track of their data. So, they
- 3:46are losing products that were otherwise
- 3:48good to be sold.
- 3:50Currently, their losses are at 1% of
- 3:52revenues, and that's cost of product
- 3:55lost divided by total revenues.
- 3:57While their main competitors are at
- 4:000.5%, and they've hired you to help them
- 4:03reverse those losses.
- 4:04So, how can DSC reduce their losses?
- 4:08And now to the follow-up question,
- 4:10the hypothesis testing question.
- 4:12You mentioned it's possible that poor
- 4:14handling of products in distribution
- 4:16centers
- 4:17is a possible cause
- 4:19uh
- 4:20for the losses. How would you test
- 4:23whether that is true or not?
- 4:25You should know that we have two
- 4:27different distribution centers, and each
- 4:29one of those serves roughly half of our
- 4:31stores.
- 4:32All right, so take your time, come up
- 4:34with your answer, but definitely come up
- 4:36with your answer again after hearing
- 4:38each one of the steps, so that you can
- 4:40learn each one of them thoroughly, and
- 4:42then apply them in your practice from
- 4:44then on.
- 4:45Now, I'm going to explain step one of
- 4:48the three step process, and then I'm
- 4:50going to show you how I would do step
- 4:52one, and you should probably do step one
- 4:56before watching my answer, and then
- 4:58watch it and try to understand what's
- 4:59different, and only then move on to step
- 5:01two. Okay?
- 5:03So,
- 5:04the way I
- 5:05the what you need to know is
- 5:08defining your hypothesis means
- 5:11defining what is the cause, the effect,
- 5:12and the consequence in this hypothesis.
- 5:14And all hypotheses must have a cause, an
- 5:17effect, and a consequence. I'll show you
- 5:19an example of this in a second. And next
- 5:22thing you need to know is this step is
- 5:23only here to help you. You might decide
- 5:26to
- 5:28do it and show it to the interviewer,
- 5:29you might decide to
- 5:31do it and not show it to the
- 5:32interviewer, and you might even not
- 5:34decide you might even decide not to do
- 5:36it at all, and that does not mean you're
- 5:38not going to pass this question, you're
- 5:40not going to do well in this question.
- 5:42But, this step is meant to help you
- 5:44clear your thoughts and come up with
- 5:46your test, and it really, really does
- 5:48help it. Okay? So,
- 5:51when you break up your hypothesis like
- 5:52this, you have
- 5:56much more clarity of what is there, what
- 5:58is it you want to test, what is it you
- 6:00can test, and what you can't. Okay? Let
- 6:02me show you an example so this is going
- 6:03to be more clear.
- 6:05Uh so, the question is,
- 6:08is poor handling of products in
- 6:10distribution centers a cause the cause
- 6:12for the losses?
- 6:14Mm the
- 6:16cause, effect, and consequence here are:
- 6:19Cause, distribution center employees are
- 6:21poorly trained or reckless, and that
- 6:24causes them to
- 6:27handle products poorly,
- 6:29and the consequence of that is that
- 6:31distribution centers cause losses of
- 6:34otherwise fine products. And
- 6:38even if you decide not to show this
- 6:39first part to the interviewer, the
- 6:40second part of this step is crucial, and
- 6:43I would definitely recommend that you
- 6:44show it to the interviewer, which is
- 6:46stating the sentence of the hypothesis
- 6:48you want to test. And the one here would
- 6:50be, the excessive product losses faced
- 6:53faced by drugstore chain DSC are a
- 6:56result of bad handling in either one or
- 6:59both of their distribution centers.
- 7:02Okay? So,
- 7:04let's move on to step two, breaking the
- 7:06hypothesis down into simpler hypotheses.
- 7:09And the logic behind this step is
- 7:11breaking your hypothesis down into
- 7:15those other hypotheses that need to be
- 7:17true in order for that one to be true.
- 7:19Uh you need to do that because
- 7:22you need simple hypotheses to test. You
- 7:25can't test overly complicated
- 7:26hypotheses.
- 7:28Uh
- 7:29if your hypothesis is simple enough,
- 7:30though, when you can test it already,
- 7:32then you don't need to do that. How do
- 7:33you know if it's simple enough?
- 7:35Again,
- 7:36it's only simple enough if you can test
- 7:38it straight away. If you can't test it
- 7:40straight away, then you probably need to
- 7:42break it down. Okay? So,
- 7:45the first principle that you need to
- 7:48know here is that logical breakdowns are
- 7:50better. Your breakdown might not be 100%
- 7:53MECE, but the closer to that the better.
- 7:56And you might not test all of the
- 8:00sub-hypotheses, the simpler hypotheses
- 8:01that you come up with, but you need to
- 8:03at least show your interviewer that you
- 8:05have a plan
- 8:07of what you're going to do if you need
- 8:08even more certainty. Okay? And this
- 8:12brings me to the second principle, quick
- 8:14tests versus exhaustive tests. So, if
- 8:17you need to run quick tests, you can
- 8:19always pick the one or two most
- 8:20important ones and controversial ones
- 8:24and uh in your breakdown. So, that means
- 8:28that you'll pick the ones that are going
- 8:30to mean the most to the main hypothesis
- 8:33and that you know the least the answer
- 8:36to.
- 8:38So, if it's something quite simple or
- 8:39that you have a good idea of what it's
- 8:41going to be, then that might not be the
- 8:43test that you want to do straight away.
- 8:45Uh and if it's something that's not
- 8:47going to affect so much the main
- 8:49hypothesis that you're trying to test,
- 8:51then you might not want to test it
- 8:52straight away, too. You might want to
- 8:54pick one that you don't know the answer
- 8:56to and that it's really going to change
- 9:00the chances that the main hypothesis is
- 9:02true or false.
- 9:04Okay?
- 9:05And
- 9:06uh this is not the only way to do
- 9:08quicker tests. Uh we're going to see
- 9:10this in the third step, but it's one of
- 9:13the ways and you're probably going to
- 9:14use both of them together. Next, uh
- 9:17availability of data should be in your
- 9:19mind when you're doing this step. You're
- 9:21not yet going to tell the interviewer
- 9:23how you're going to come up with the
- 9:25with the data, how you're going to
- 9:26gather that data, and how you're going
- 9:27to process it, but you don't want to put
- 9:30yourself in a dead end. So, if
- 9:34you obviously can't get data for some of
- 9:36the most important tests uh in your
- 9:39breakdown, then you probably should
- 9:42try to come up with a different way to
- 9:44break down this problem because
- 9:46otherwise you're going to put yourself
- 9:47in a dead end and you're not going to be
- 9:48able to complete your hypothesis test.
- 9:51And now I'm going to show you my answer
- 9:53or at least the second step of my
- 9:54answer. So make sure you've done it
- 9:57yourself too before you watch mine.
- 10:00So the hypothesis I want to test here is
- 10:03are excess losses caused by poor
- 10:05handling in distribution centers?
- 10:07And by excess losses, I mean those
- 10:11losses that are above 0.5% of revenues,
- 10:15which is my benchmark. So there are two
- 10:17main hypothesis that will answer this
- 10:19question. First one is do these excess
- 10:21losses really come from distribution
- 10:23centers? And the second one, are they
- 10:25really caused by mishandling? If they
- 10:27come from distribution centers and are
- 10:29caused by mishandling, then this
- 10:30hypothesis has to be true.
- 10:32And to find out if they come from
- 10:34distribution centers,
- 10:35I would try to figure out where in the
- 10:38product's path from supplier to consumer
- 10:41do these excess losses come from? And in
- 10:43a second I'm going to talk about how I
- 10:45would find that out.
- 10:47And to find out if they're really caused
- 10:49by mishandling, first I'd need to know
- 10:51what kind of loss is behind the excess
- 10:54losses. So are they caused by
- 10:57expired products? Are those products
- 10:59going missing? Or are those products
- 11:01damaged? Number two, where in the
- 11:04distribution center process do these
- 11:06losses come from? So
- 11:08uh from supplier to leaving to the
- 11:10store, leaving distribution center to
- 11:12the store, there are several steps in
- 11:14the distribution center process. I want
- 11:16to know where it is and then I want to
- 11:18investigate that
- 11:20process that step of the process to
- 11:22figure out what exactly is behind that.
- 11:26Okay? So
- 11:28now let's move on to step three. I'm
- 11:30going to show you how it works and then
- 11:32I'm going to show you
- 11:33um the last step of my answer.
- 11:36The third step is how you would gather
- 11:40and process data. That actually means
- 11:43what you would be doing in real life.
- 11:45This is what we want to tell the
- 11:46interviewer. So, what data do you want
- 11:48to get? How would you get it? How would
- 11:50you turn it into information? What do
- 11:51you expect it to look like?
- 11:53Again, of course, you might not say all
- 11:55of these things to the interviewer, but
- 11:58you at least need to show them that if
- 12:00you were in a real project, if this was
- 12:03was your assignment, you'd be able to
- 12:06complete it well and it'd be clear to
- 12:09your manager, to the partner, to the
- 12:11client uh whether the hypothesis is true
- 12:14or not and you'd be based on data to do
- 12:16that and of course you can get that data
- 12:19and process it yourself.
- 12:21And next
- 12:23indicative tests versus conclusive
- 12:25tests. As I said before,
- 12:27uh choosing only a part of the
- 12:30hypothesis to test is not the only way
- 12:32to do it quicker. You can do indicative
- 12:34tests uh which are tests that don't
- 12:38necessarily prove your hypothesis to be
- 12:41false, but shows that it's much more
- 12:43likely for it to be true or false versus
- 12:46doing conclusive tests where the data
- 12:47data actually prove the hypothesis or
- 12:50very very very close to proving the
- 12:51hypothesis. I'm going to show you some
- 12:53examples of this in my own answer. Okay?
- 12:57So,
- 12:58take your time and come up with the
- 12:59third step of your answer or with the
- 13:01full answer all over again if you want
- 13:03to and only then
- 13:05uh watch my answer which I'm going to
- 13:07show you now.
- 13:08Uh first thing, well, here is my second
- 13:12step
- 13:13and I'm going to I want to show you how
- 13:15I would test the first part of my little
- 13:20framework here. And this is
- 13:23I tried to draw a chart like this. All
- 13:25the steps from supplier to consumer. So,
- 13:28arrival from the supplier, distribution
- 13:30center, transport, and store. And I want
- 13:32to compare my losses with my benchmark
- 13:35in each one of those steps. What I
- 13:37expect it to look like is
- 13:40I have a lot more losses in the
- 13:41distribution center than my benchmark.
- 13:44And the losses are similar in all other
- 13:47steps.
- 13:48If I can't get data from my benchmark,
- 13:51which is likely, I would test one
- 13:54distribution center against the other.
- 13:56Uh so, I'd get my losses as a percentage
- 13:59of revenues
- 14:00in each one of the steps. And what I
- 14:02would expect it to look like is one
- 14:04distribution center having much more
- 14:06losses than the other. Or else,
- 14:09uh both distribution centers with much
- 14:11more losses than all other steps.
- 14:14So,
- 14:15take a look at this.
- 14:17These second tests are not conclusive.
- 14:21Uh so,
- 14:23benchmarking one distribution center
- 14:25against the other does not conclude to
- 14:27me, even if one of these Actually, even
- 14:30if this last one is the case, it doesn't
- 14:33prove to me that distribution center is
- 14:36the cause. Maybe all of the losses are
- 14:38supposed to come from distribution
- 14:40centers. And what's weird in the FC is
- 14:44that we have losses in arrival from
- 14:45supplier, we have losses in transport,
- 14:47we have losses in store, or something
- 14:49like that. And we wouldn't know because
- 14:51we can't compare it to the benchmark.
- 14:53Which might even have more losses than
- 14:55us in the distribution centers, but not
- 14:57so many so much more losses. But no no
- 15:00losses at all in the other steps. Okay?
- 15:03But this, at least, indicates that we
- 15:06have a problem in distribution center,
- 15:08indeed.
- 15:10And another lesson here is that anytime
- 15:13you have the opportunity to draw charts,
- 15:15and that would make communication
- 15:16simpler,
- 15:17take that chance and do it. That's going
- 15:19to help you speak the interviewer
- 15:21language, language. Communication is
- 15:22going to be much more much simpler and
- 15:25much much clearer.
- 15:26And also, something that I want you to
- 15:29notice here is there's additional
- 15:31information in this chart that would
- 15:33have been hard, or at least not as quick
- 15:36to convey if I were speaking it only.
- 15:38And this is what are the steps in the
- 15:40distribution center uh
- 15:43in the process from the consumer to the
- 15:44supplier. So,
- 15:46whereas before I only said the steps
- 15:49from supplier to consumer, now I'm
- 15:51showing what would be the steps that I
- 15:54would use in my analysis.
- 15:56And next, what I expect the data to look
- 15:59like. So,
- 16:00I drew I clearly drew here
- 16:02the my expectations of
- 16:05uh
- 16:06what the data should look like. If it
- 16:08does not look like this, then my
- 16:09conclusion would be
- 16:11slightly or a lot different.
- 16:13Now, let's move on to the next step
- 16:15where we're to the next uh sub
- 16:17hypothesis, where you're going to show
- 16:20where going to we're going to see all of
- 16:21this again.
- 16:22So, what kind of loss is behind the
- 16:24excess losses? How am I going to find
- 16:26that out? I want to break down those
- 16:28losses between those three types that I
- 16:30said before.
- 16:31Uh missing products, expired products,
- 16:33and damaged products. And I want to
- 16:35compare distribution centers if I can't
- 16:37compare it to benchmarks. And I believe
- 16:40it's highly unlikely that I can't
- 16:41compare it to benchmarks. If I can,
- 16:43well, that's great. If I can't, I'm
- 16:45going to do this indicative test here
- 16:47and have an idea of what's going on. And
- 16:50what I expect to see is
- 16:53either uh
- 16:55a lot more damaged products in
- 16:57distribution centers
- 16:59or in only one of them.
- 17:01Okay? And to find out where in the
- 17:04distribution center process these losses
- 17:07come from, I would divide
- 17:09I would do pretty much the same as I did
- 17:12in the
- 17:14in the first test.
- 17:16Draw out the process, but I would find
- 17:20the losses by type of loss in each step
- 17:23of the process. So,
- 17:25what losses are
- 17:27expired products, missing products, and
- 17:29damaged products in the reception, in
- 17:31sorting, in storage, in picking, in
- 17:34boxing, and in exit. And if I see
- 17:36something like what I'm seeing in this
- 17:38chart, which is
- 17:40a lot of damaged product losses, which
- 17:42are much more than all other losses in
- 17:45one or two steps, or even all of them,
- 17:48then
- 17:49there's probably something going on, and
- 17:51I want to go invest and investigate the
- 17:54root cause of that to find out that is
- 17:57actually it actually is
- 17:59uh
- 18:00mishandling. And the way I do that is
- 18:03probably go to the distribution center
- 18:06or to the distribution centers,
- 18:08and
- 18:09look at the process, see what's going
- 18:11on, try to see the damaged products, try
- 18:13to see them being damaged. I would talk
- 18:16before that probably to the manager and
- 18:19to whoever is in charge of that specific
- 18:22process, and try to get try to gather
- 18:25from them uh
- 18:26what they think is causing those losses,
- 18:29what is different from the other
- 18:31distribution center in there, and then
- 18:33I'd get to the root cause of the
- 18:35problem.
- 18:36And now,
- 18:37to tell you, like I said in the
- 18:39beginning,
- 18:40an interesting way to come up with your
- 18:43own hypothesis testing questions, it's
- 18:45fairly simple. So, you've seen in
- 18:47frameworks before
- 18:49that you
- 18:51are always going to come up with several
- 18:52hypotheses by the end of your framework.
- 18:55Uh and
- 18:57the interviewer could pick any one of
- 18:59those and ask you how you would test
- 19:01them.
- 19:02Uh and what you can do is
- 19:05draw your framework for any case,
- 19:08check look at what your hypotheses are,
- 19:11and each one of those can work as a
- 19:14follow-up question for hypothesis
- 19:15testing. Uh pick any one of those or
- 19:18pick all of those and test them and that
- 19:20is your way of practicing. As I said,
- 19:22practicing is the only real way to uh
- 19:26learn hypothesis testing. If you do that
- 19:29with partners, not necessarily one
- 19:30interviewing the other, but both come up
- 19:33with your own answers and then you
- 19:34discuss both of them, which one is
- 19:38more
- 19:39which one will give you more certainty
- 19:41of your conclusion, which one won't,
- 19:43which one's quicker or less costly, and
- 19:45which one is less.
- 19:47And
- 19:48you're going to have you're going to
- 19:50uh
- 19:52you're going to be able to practice and
- 19:54learn a lot with that. Okay? So, I hope
- 19:56this tool helped you. I hope this theory
- 19:59tool helped you and I'll see you in the
- 20:01next videos. You're going to have
- 20:03several
- 20:04different hypothesis testing videos to
- 20:06help you here and I honestly hope you
- 20:09can learn this and apply this and
- 20:10improve your case performance and
- 20:12hopefully get your offer. See you in the
- 20:14next one.
About this transcript
This page contains the full transcript of Case Interview Fundamentals: Lesson 29 [Hypothesis Testing: the 3-step approach] by CraftingCases, generated from the public captions YouTube serves with the video. The transcript has 3,413 words across 555 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 choose a caption track available on the video.
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, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.