YT YouTube2Text

YouTube2Text → Transcripts → BREAD-IGC virtual PhD-level course 2026 | Lecture 2 History, Persistence, and Development

BREAD-IGC virtual PhD-level course 2026 | Lecture 2 History, Persistence, and Development — Transcript

by International Growth Centre · 14,592 words · 2,193 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:06Oh, and if we respond, do we respond to
0:08everyone in the chat or hosts and
0:11panelists?
0:12>> Um, most the questions will come through
0:14the Q&A box.
0:17>> Oh, not through the chat.
0:18>> Not through the chat, right?
0:20>> Okay. Q&A box is somewhere else.
0:23>> Yeah,
0:24it's down there as well.
0:28Hi everyone. I just want to give you a
0:30warm welcome to our second week of the
0:33bread lecture series for political
0:35economy and development. Um, we'd like
0:38to really uh thank Nathan Nun, who uh is
0:43a professor at the University of British
0:45Columbia, and Melissa Dell, who is a
0:48professor of economics at Harvard for
0:51joining us today to teach us um a bit
0:54more about history, persistence, and
0:56development. And um before I I hand over
1:01to them, just quickly want to remind
1:03everyone uh if you have any questions,
1:06please put them in the Q&A box. We love
1:09introductions, but we we'd love for the
1:11Q&A box to be around more um content
1:15related questions.
1:17The second thing is that the IGC website
1:20I will share again is where all the
1:22lectures, slides, recordings, and
1:24reading lists will be. Uh so please
1:28refer to that if you have questions.
1:31>> And the last thing is um if you would
1:35like to be a panelist and have your mic
1:38and microphone or your your microphone
1:40and video enabled, please raise your
1:43hand, but only if you're actually going
1:45to have your your video on. The idea is
1:46that this is kind of like a classroom
1:49interaction. So um if your video won't
1:52be on, then then I'll just kind of
1:54switch you out with someone else. But
1:56yeah, raise your hands now and I'll
1:58randomly assign 10 20 people and we'll
2:01go from there.
2:02>> I never was.
2:03>> Okay. Um Melissa, over to you.
2:07>> All right. Awesome. Uh thanks so much
2:08for the introduction.
2:11Let's
2:13All right. Um so research on the long
2:15run determinance of economic growth, I
2:18would say, has really transformed
2:19economic history and even economics more
2:22generally over the past 25 years. Um, so
2:24it's helped inspire the integration of
2:27modern empirical methods into economic
2:29history. Um, and I think it's also
2:31contributed to the expansion of
2:33economics beyond its traditional focus
2:35on the US and Europe. Um, insights from
2:38this literature have the potential to
2:40inform policy um by helping us to better
2:43understand how policy interacts with the
2:45broader context and how its effects
2:46unfold over time.
2:49And so as Bob Solo said, the part of
2:51economics that's independent of history
2:53and social context is not only small but
2:55dull. Um so hopefully that means that
2:57today won't be dull. And so just to you
3:00know to start with a bit of intellectual
3:02history. Um if we go back to the 1940s,
3:04the 1950s, economic history and
3:07development economics were very central
3:10um to to economics more generally. um
3:13writings in this genre tended to be
3:15really rich in institutional detail and
3:18historical detail, but they were
3:19typically not disciplined by a formal
3:21model. Of course, they also weren't
3:23disciplined by data given that uh there
3:25were no computers at that time. Um by
3:28the 1960s, economics journals
3:30essentially wouldn't publish anything
3:32that was not disciplined by a
3:33mathematical model. um and economic
3:36history, development economics really
3:38moved um outside the mainstream of uh
3:42[clears throat] traditional economics.
3:43Um most mathematical models of economic
3:46growth traditionally had a single long
3:48run steady state and in those models
3:50history is irrelevant. Uh development
3:53practitioners following the solo solo
3:55model tended to emphasize the
3:57accumulation of capital first physical
3:59capital and later human capital. Um and
4:02economic historians who were interested
4:05in development tend to focus on the
4:08historical ascendancy of the US and of
4:10Europe. And so along with the resurgence
4:14of development economics in the 1990s
4:16and 2000s, there was really a renewed
4:18interest in the role that history plays
4:20in economic growth.
4:23And so a sort of a a fun analogy comes
4:25from Paul Krugman um in talking about
4:28kind of the decline and then renaissance
4:30of of development economics where he
4:32compares it to um African maps. And so
4:35he shows first of all a map from the
4:371550s. And so you can see this is very
4:39rich in detail but it's all completely
4:41made up. Uh people essentially didn't
4:43know uh what what they didn't know. uh
4:45people uh European mapmakers if you go
4:48forward to 1805 by that uh point
4:51European map makers kind of understand
4:53what they don't know and most of the map
4:55is actually empty because they say well
4:56we have no idea uh what's here and then
4:59as we move uh further forward in time
5:02they start to fill in the details and
5:04the argument is kind of development
5:05economics um is is has been similar and
5:10I think one of the insights that's uh
5:12came from this past you know 25 years of
5:15trying to fill in this map is that the
5:17channels of persistence um matter.
5:20Historical events or historical
5:22institutions that at first glance appear
5:24to be similar can have very different
5:27long run effects which doesn't apply
5:29that you know it doesn't mean that we
5:30can't generalize but it suggests that
5:33it's important to consider the
5:34transmission mechanisms. And so we want
5:36to move beyond just emphasizing the
5:38treatment, you know, for example,
5:40extractive institutions and towards an
5:43examination of which intervening
5:46variables matter given a particular
5:48context. And so there's kind of many
5:50illustrations of this. I'm going to give
5:52one um brief illustration uh with some
5:55of my work um uh about extractive
5:59institutions. And so we're going to
6:00briefly consider two different
6:01extractive institutions. The first one
6:03is the Peruvian uh mining mita. Um and
6:07the second one is the cultivation system
6:09in Java which draws upon uh work that
6:11I've done with uh Ben Olen.
6:14And so the mining mita was a forced
6:18labor system instituted by the Spanish
6:20colonial government in the 16th century.
6:22It persisted for over 200 years. And it
6:25required indigenous communities in
6:27Highland, Peru, and Bolivia to send 17th
6:30of their adult male population to work
6:32in uh silver and mercury mines. Um, and
6:36so the Potosi mines provided the largest
6:38deposits of silver in the Spanish
6:39Empire. You know, if you've heard of
6:41Spain extracting precious metals uh from
6:44their colonies, a lot of that is coming
6:45from Potosi. Um,
6:48and this is a picture of um the the
6:51Mita's extent. And so it's essentially
6:52anywhere kind of inside that um you know
6:55that that boundary in highland Peru um
6:58is is contributing uh people to the
7:00mita. Um an alternative extractive
7:03institution is the cultivation system in
7:06Java. Um so the island of Java and
7:08what's today Indonesia was the center of
7:10the Dutch colonial empire. Um from the
7:13early 1830s through the 1880s the Dutch
7:16colonial state forced people along
7:18Java's northern coast to cultivate
7:20sugar. um which was then processed um in
7:24colonial sugar factories for export in
7:26Europe. So the revenues at their peak
7:28accounted for over one-third of Dutch
7:30government revenues and over 5% of GDP.
7:33In monetary terms, this is the largest
7:35example of colonial extraction um in in
7:39history. Uh ultimately sugar production
7:42collapsed in the Great Depression and
7:44Indonesia is now the world's largest
7:45sugar importer.
7:48And [snorts] so raw sugar cane is very
7:50very heavy. You can't transport it very
7:52far. And so the Dutch uh established 94
7:55watered sugar factories um along um in
7:59in northern Java um and the sugar had to
8:02be grown nearby. So there was
8:04essentially a catchment area around each
8:06uh factory um and um people living
8:10inside that catchment area uh were
8:12forced to reorganize their land to grow
8:14sugar cane. and uh local officials uh
8:18local you know uh Javan officials within
8:20the catchment areas were empowered by
8:22the uh Dutch colony to make this happen.
8:27Okay. And so this is a map of of the
8:29cultivation system. It's a bit small but
8:31the black dots are sugar factories and
8:33then that you see the catchment areas
8:35around each of the factories.
8:38All right. So what were what were the
8:40impacts of these uh two different
8:42prominent examples of colonial
8:44extraction? Um well in the case of the
8:48Mita um we find that it lowered long run
8:51economic development. Why would that be
8:53the case? Um well what I try to show
8:56kind of with various data in the paper
8:58is that um importantly the Spanish
9:01colonial uh government would not let
9:04Spanish land owners settle in the
9:07districts that had to contribute forced
9:09labor to the mines. Why why wouldn't
9:11they do that? Well, it was competition
9:12for labor. if a landowner was trying to
9:16exploit labor and then the state comes
9:18around and says we need this labor to
9:19work in the mines um you know that would
9:22create a conflict and the colonial
9:23government didn't want that conflict so
9:25they didn't let Spanish land owners into
9:26these communities that contributed uh
9:29forced mining labor for the state um and
9:33so you had the large land owners settle
9:35outside and that's very persistent
9:37across time and so in this particular
9:40context the alternative to large land
9:42owners was not relative equality between
9:44small infranchised land owners. You
9:46know, the Mita districts, they don't
9:47look like colonial New England. Um
9:50instead, the alternative was poorly
9:52defined property rights and the
9:53underprovision of public goods. Um so
9:56certainly the areas kind of the non-mea
9:58districts, the areas that didn't have to
9:59contribute the mining labor were not
10:01organized to benefit uh the masses, but
10:04the evidence suggests that large land
10:06owners who were located there served as
10:08a stabilizing force. um lobbying for
10:11public goods provision that benefit the
10:13general populace today kind of long
10:15after land reform uh now that the land
10:17owners are are gone. Um the cultivation
10:21system in contrast we sure that it had
10:24positive long run effects on local
10:26economic development. Um and this is
10:28true both for places that are closer um
10:31to the factories as well as for the
10:33communities that were forced to
10:35contribute um uh labor to cultivate uh
10:38sugar. And so why would this be the
10:40case? Um well, we find various evident
10:43that economic changes induced by the
10:45cultivation system led to a more modern
10:47economy with more manufacturing, greater
10:50retail, greater education levels. This
10:53happened both near the factories and in
10:55subjected villages where you know
10:57following the abolition of the
10:58cultivation system the local government
11:01retained control of the land that had
11:03been used to cultivate sugar cane and
11:05use the proceeds from that land um to to
11:08pay for teachers in some cases. And so
11:10you see education actually uh being
11:12higher. And so they kind of took this
11:14system initially designed to kind of
11:16extract a s surplus for the colonizer
11:19and then took that land and put it
11:21towards providing local public goods in
11:23this case education. So these economic
11:26changes that were um instituted uh for
11:30the cultivation system so that the Dutch
11:32could create the surplus that they
11:33wanted to extract persistently influence
11:36the development trajectory through
11:38investments in infrastructure. The Dutch
11:40built many roads and railroads to the
11:42sugar factories to take the sugar out
11:44and those persist to today through input
11:46output linkages through elomeration and
11:48through human capital accumulation.
11:52Okay. So in short why does it appear
11:54that these two prominent examples of
11:56colonial extraction have kind of very
11:58different effects in the long run. Uh
12:02well one important difference is the
12:04role of manufacturing. The cultivation
12:07system required substantial local
12:09manufacturing in order to process the
12:11sugarcane and linkages between sugar and
12:14other sectors plausibly provided a key
12:17uh mechanism through which
12:19industrialization propagated and
12:21persisted. In contrast uh with the
12:23mining mita, the conscripts were
12:25marched, you know, hundreds of
12:28kilometers from their communities to a
12:29single location to mine silver that was
12:32then exported, you know, minimal need
12:34for local processing. We could have more
12:36examples of this. So if you think about
12:39um uh the uh Congo um where evidence um
12:43from Lowe's and Montero has also showing
12:45kind of very negative longrun effects of
12:48colonial extraction. Well, there you
12:50just needed to cut the rubber and dry it
12:52before it could be exported to Europe.
12:54So again, no need to build a
12:55manufacturing infrastructure. You know,
12:58another difference is investment in
13:01infrastructure more generally. Um, and
13:03so the Dutch made major infrastructure
13:06investments in both rail and roads to
13:08transport the manufactured sugar to the
13:10ports. And if you look at where roads
13:13and railroads are today in Java, they're
13:15in the same place that the Dutch, you
13:17know, built them uh well over a century
13:19ago. Um in the Peruvian case, you know,
13:22you just need to carry the silver on a
13:24mule down to the coast. There's not
13:26infrastructure built. And in fact, the
13:27infrastructure that was later built
13:29ended up disproportionately outside the
13:31Mita areas where you had large land
13:33owners who could lobby for it. I mean,
13:35again, we could talk about the Belgian
13:37example or other examples. um you know
13:39there was a strong incentive in that
13:42case to maximize shortrun instruction um
13:45with again very little infrastructure
13:46that was provided.
13:49Okay, so this is just one example. I
13:52think the idea that context matters,
13:55that the mechanisms of persistence
13:57matter is not in itself highly
13:59controversial. But often times, you
14:02know, people may not be able to kind of
14:05explore the context or the channels of
14:08persistence to the extent that we would
14:10like. Um because you have a data
14:13constraint. uh we oftentimes lack
14:16comparable granular data across time or
14:19across diverse contexts which means we
14:21often have to focus you know in a
14:23particular time period in a particular
14:26kind of narrowly defined region um and
14:28so in short measurement is central and
14:31for this reason measurement is where
14:33I've kind of put much of my attention um
14:36and I want to spend the rest of our time
14:38with just sort of a broad overview of
14:42measurement kind of how we think about
14:44measurement, where it's going and how it
14:46has the potential to transform uh the
14:49study of persistence and of historical
14:52economic development um in the coming
14:54years.
14:57Okay. So to talk about measurement, I
14:59want to go back to the maps analogy that
15:02we had in the beginning. Um and in
15:05particular I want to emphasize a short
15:08story uh by War Luis Boures um who
15:11imagines cgraphers who pursue the
15:15perfect map um until they produce a map
15:19whose size was that of the empire. And
15:23in this story this perfect map is
15:26useless uh because it no longer reduces
15:29complexity. It ceases to function as a
15:32map. Um, and
15:35measurement in many ways is kind of
15:37similar to this map. We already have a
15:40perfect representation of reality, which
15:42is just reality itself. But reality is
15:45incredibly complex, highdimensional. You
15:48know, we can't put it into a regression
15:50analysis. It's just too complex for us
15:52to make sense of it. And so, we need a
15:54measurement. We need a way to simplify
15:56reality that provides a map um to to
15:59whatever question to whatever
16:01conclusions that we're trying to
16:02examine. And so in a formal sense we can
16:05think of measurement as a projection and
16:08it specifies and implements some
16:10function that maps a highdimensional
16:13reality into lower dimensional measures
16:16uh that can be interpreted and that can
16:18be analyzed. uh because reality you know
16:22which we call X here is very
16:23highdimensional there may be many
16:26different ways to measure a given
16:28concept you know what exactly does trust
16:30mean trust is a you know a
16:33multi-dimensional latent concept there's
16:35potentially many different ways you
16:36could measure it um in these kind of
16:38different measurement functions these
16:40different ways of simplifying reality
16:42preserve different features and hence
16:44they may support different empirical
16:46conclusions
16:48um in particular particular, I'd like to
16:51kind of emphasize three different stages
16:53of measurement. Um, you know, we often
16:55think of it as just, you know, we're
16:57passively observing something using an
16:59instrument or a procedure, but really if
17:01we think about it end to end, there's
17:03more to it than that. Um, and so the
17:05first stage of measurement is discovery.
17:08Given the complexity of the world, what
17:10features even exist in in the first
17:12place and what's worth measuring for a
17:14particular question? The second stage of
17:17measurement is construct definition. So
17:19once we determine the kind of the broad
17:21concept that we would like to measure,
17:23how do we actually operationalize this
17:25in a specific setting? And then finally
17:28is observation. Now we have a precise
17:31def observable definition of what we
17:33want to measure. How can we actually
17:35implement that at scale? And each stage
17:38of measurement involves different types
17:40of uncertainty and has traditionally
17:42been examined through different
17:43methodological lenses.
17:47And so in discovery, what patterns, what
17:49dimensions, what relationships exist and
17:52are worth measuring? Traditionally, we
17:53approach this through theory, through
17:56contextual knowledge. Um, but
17:58increasingly, there's new methods that
18:00would allow us to use datadriven
18:02discovery, and I'll speak about that um
18:04briefly and the potential advantages of
18:07that in certain settings. Construct
18:10definition. How should a phenomenon of
18:12interest be translated into an
18:14operational definition for a given
18:16setting? Again, traditionally, we'd use
18:18theory or domain expertise to design,
18:21you know, what does trust mean in this
18:23particular setting. Uh we might use
18:25piloting. Um we might just have to make
18:27do with whatever externally available
18:29proxy exists. Um and there's formal
18:32literatures on this mostly mostly
18:34outside of economics. Um finally,
18:37observation. How accurately can the
18:39construct be observed at scale and how
18:42do errors in measurement propagate to
18:44the parameters that we want to estimate?
18:46And so this is really the domain of
18:48measurement error.
18:52So traditionally um we typically treat
18:55measurement as fixed. Um we often have
18:58to rely on some imperfect but available
19:01proxy um as a fixed measure of something
19:05that that that we'd like to analyze, you
19:07know. So, we know that expropriation
19:09risk for foreigners is not a perfect
19:11measure of property rights. But if
19:13that's the only measure available to
19:15study that question, um, then really the
19:17focus is on justifying that it's
19:19informative enough about what we
19:21actually care about to make progress.
19:26AI is really changing kind of this this
19:30bottleneck though. So traditionally, you
19:33know, especially in areas like
19:34development history, we just want to
19:36find one measure um one proxy for what
19:40we care about and that itself can be
19:42kind of very expensive, very costly. Uh
19:45but at their core, deep neural networks,
19:47which are the foundation for things like
19:49large language models, are flexible
19:52measurement functions. They take
19:53highdimensional data which is like raw
19:56text raw image pixels and they map that
19:59into some lowdimensional measure that
20:01you can analyze. So for instance they
20:03might say you know is the sentiment of
20:05this text positive or negative. Say take
20:08the highdimensional text map it to 01
20:11measure of if that text has a particular
20:13sentiment.
20:15Traditionally
20:17uh researchers would use handgineered
20:19rules to extract information from text
20:22or from images. Neural networks which
20:25again are the kind of the the foundation
20:27of modern AI instead learn
20:30representations from empirical examples.
20:32So in short they use millions to
20:34billions of parameters to transform
20:37highdimensional inputs. So raw text or
20:39images into lower dimensional measures
20:41that could be a label. What's the topic
20:44of this text? that could itself be a
20:46continuous uh vector. Um, and they are
20:50trained to perform specific tasks. Um,
20:53and with sufficient training data and
20:55appropriate objective, they learn
20:57features of reality that are useful to
20:59preserve for the task. Part of the
21:01reason why they're so powerful is that
21:03they can generalize to an extent beyond
21:06the task on which they're trained.
21:08Across many tasks, they outperform human
21:10rules. And they're highly scalable. AI
21:13can substantially lower the costs of
21:15converting unstructured data, so text or
21:18images into lowdimensional variables for
21:20statistical analysis. And we already see
21:22this kind of starting to transform um
21:25economic history and the study of long
21:27run development. I want to emphasize and
21:31I think this is really important that in
21:33many settings credible AI measurement
21:36requires substantial fix investments.
21:39You know, so when I gave this talk a few
21:40years ago, I would give lots of examples
21:43of all the things you can do with neural
21:44networks and why it's so cool and how
21:46everybody should be using this. Um, you
21:49know, now if anything, I think we've
21:50swung to the opposite extreme where
21:52people are using kind of off-the-shelf
21:54LLMs to do everything. Um, it's very
21:56important to carefully validate your
21:59data. you know, in in trying to digitize
22:01lots of documents, oftentimes the the
22:04layouts are wrong, like right, it puts
22:06the address with the wrong name on the
22:08page. It still struggles to align the
22:10data, you know, so that everything's in
22:12the right row in the right column. It
22:13hallucinates like crazy. You know, I
22:15accidentally feed in a page upside down
22:18and it keeps on predicting, you know,
22:20names from the same ethnic group,
22:21addresses from the same neighborhood,
22:23but completely made up because it
22:24doesn't know how to read my upside down
22:26page. So in a lot of contexts, you know,
22:29historical data that's very different
22:31than the web text that uh large language
22:34models are trained on, uh, you know, in
22:36in low resource settings or languages,
22:38you still need a lot of customization.
22:41So it doesn't mean measurement is always
22:42cheap, but at a minimum, it's cheap to
22:44scale once you have a good model. And in
22:47some kind of relatively straightforward
22:49tasks, um, the off-the-shelf model uh
22:52can do relatively well. Um, and so when
22:55we're in this world, it makes it
22:57inexpensive to construct many different
23:00measures of the same broad concept. And
23:02so in short, the bottleneck shifts from
23:04having any scalable measure of what
23:07you'd like to study to choosing amongst
23:10many plausible measures which may yield
23:12different estimates of your target
23:14parameter.
23:16Um, these challenges closely parallel
23:18those created by the uh rise of personal
23:21computing in the 1990s. And so, you
23:24know, back then, cheap compute enabled
23:26researchers to estimate not one
23:28specification as they would have on a
23:30mainframe computer, but hundreds,
23:32thousands or even millions of
23:34specifications.
23:36Um,
23:37and so that forced us to ask what should
23:40discipline amongst many possible
23:42specification? What makes a
23:44specification credible? And you know,
23:46you had literatures on causal inference,
23:48on robustness, on multiple testing. Um
23:50an AI enabled measurement is very much
23:52moving this specification search problem
23:55upstream by expanding the set of
23:57feasible measurements. It underscores
24:00the importance of assessing what makes a
24:02measure credible and how to choose
24:05amongst many possible measures. Um and
24:08you know it can expand possibilities at
24:10at at all stages of measurement. Um so
24:14in the context of discovery again
24:17traditionally this would be guided by
24:19theory contextual knowledge uh the
24:21existing literature and this works well
24:24when the problem is sufficiently
24:26understood for some prespecified
24:28measurement to be well motivated. Uh but
24:31the risk is if that's the only tool you
24:34have available. You might essentially be
24:36looking for the keys under the lamp
24:38post. Right? searching only for concepts
24:40that you already know how to specify.
24:44Um AI has the potential to really um you
24:48know kind of change our approach or
24:50complement our approach to deciding what
24:53to measure and what to study in the
24:55first place because essentially what you
24:57do with data driven discovery is as the
25:00researcher you choose a way to represent
25:03um the underlying data uh rather than a
25:07predefined construct. I'm not going to
25:09have time uh to explain how this works
25:11now. You can go to the MBER methods
25:13lectures from this summer and see more
25:15information. But in short, you know,
25:17you're going to let the model summarize
25:19your text or image data by identifying
25:22structure that you don't have to
25:24prespecify in advance. Um essentially,
25:28you're going to group similar
25:29observations together. Or you might
25:31learn some continuous dimension along
25:34which the observations in your data
25:36vary.
25:37um this discovered structure can then
25:40enter your measurement in two ways. It
25:42might just help you to explore the data
25:44and to understand what's there and then
25:46you specify a more precise hypothesis.
25:49Um you know, alternatively, the
25:51discovered structure may itself
25:53constitute the measurement output.
25:54Again, I won't have time to elaborate
25:56now. Um but there's more information in
25:58the the MBR methods lectures which are
26:00posted on their website. Um,
26:04and so what's an example of how we might
26:07use datadriven discovery? Uh, well,
26:10suppose um that you are pre-registering
26:13a study that you're doing and part of
26:14that study is an open-ended survey
26:16question. I mean, it seems a bit silly
26:18to have an open-ended survey question
26:20where you then pre-register what you're
26:23going to look for in that open-ended
26:24data. I mean, why not just ask a
26:26narrowly defined question then? But you
26:27still might want to tie your hands in
26:29advance. And so instead of
26:31pre-registering
26:33uh what um you're going to look for in
26:36that open-ended survey question, instead
26:38you can pre-specify a fixed coding
26:41scheme that you're going to use to
26:43analyze what patterns appear in that
26:45data. So you've still completely tied
26:47your hands, but without having to say,
26:49I'm looking for XYZ in these open-ended
26:52responses when you know that you might
26:55totally miss what's what's interesting.
26:59All right. um construct definition. Um
27:04this can be really challenging in
27:06practice. You know, traditionally
27:07economists have emphasized things that
27:09are directly observable, but that's
27:11become much less true, I would say, over
27:13the past 15 years. There's an emphasis
27:16on things like trust, on economic policy
27:19uncertainty, job quality. Um you know,
27:22we could go on about things that are
27:24multi-dimensional, not directly
27:25observable. They're closely related to
27:27other things. And so how do you know
27:29that you've actually kind of captured a
27:31meaningful representation of that
27:33highdimensional latent concept um and
27:36not simply proxied something else? Um
27:41AI itself is not going to determine what
27:44the construct definition is. Um there's
27:46a a large and really interesting
27:48literature on this in psychometrics from
27:50back in the 1950s where people thought
27:52kind of really carefully about this
27:54question. and how do we know that we've
27:56actually captured what we intend to
27:57capture? And it really emphasizes that
28:00you need to bring in theory and domain
28:02expertise to clarify what your measure
28:05should capture and predict what it
28:07should be distinct from and where it
28:09should be stable. And to the extent that
28:12measurement becomes cheaper, you know,
28:14potentially with AI, it makes it easier
28:16to empirically validate, oh, this
28:18measure of trust captures what I would
28:20expect from theory, um, and doesn't just
28:23capture some other, um, correlated
28:25measure.
28:27And finally, observation. Um, AI serves
28:30as a scalable measurement technology uh
28:33that makes it possible to do do things
28:35like classify large amounts of text,
28:37extract information from vast image
28:40collections, and more generally to
28:42construct lowdimensional data that we
28:44can analyze uh from text and images. Um,
28:48but
28:50you know, importantly, there are a lot
28:52of decisions that go into how you use
28:54AI. Even if you're using something off
28:56the shelf, you have to choose a prompt.
28:58there's a bunch of system settings. You
28:59know, we might hope that this is
29:01innocuous. There's a big literature that
29:03shows that it's not and that small
29:05differences in how you implement things
29:08can change your measurements in
29:09meaningful way. Those diff
29:11implementation differences are not part
29:13of the economic construct itself. And so
29:16the key implication is that validation
29:18is central. And again I won't have time
29:20today to talk in detail about validation
29:23but there's an article on the reading
29:25list if you go to the MBR summer
29:27institute methods lectures we talk about
29:29this um you know along with aeshwab
29:32banchin from MIT in a lot more detail um
29:35you know so again what we you know we
29:38we're not going to perfectly capture
29:39reality that's not the goal of
29:41measurement the goal of measurement is
29:42to simplify and we want these
29:44simplifications to be useful but that
29:46can be hard to assess because reality is
29:48complex Lex, it's partially
29:50unobservable. It's multic-causal. It's
29:52not easily manipulated. And so instead,
29:54what validation emphasizes is that the
29:57researcher specifies explicit
29:59measurement criterion that are
30:01observable and they can assess how well
30:04their measurements behave relative to
30:06those explicit, interpretable, debatable
30:09measurement criteria. And this is
30:11enormously important in the age of AI
30:14because the internal workings of a
30:16neural network are much too complex to
30:19directly interpret. If you open up the
30:22neural network, it will have billions of
30:24parameters that interact in complex
30:26ways. It's not expressed as auditable
30:29rules. So you can't directly interpret
30:31it. But what you can do instead is
30:34evaluate its outputs against criteria
30:36that you the researcher have specified
30:38that are directly observable. And this
30:41is important because you know
30:43researchers have also shown that the
30:45biases of neural networks shift in
30:47complex ways and humans are really bad
30:49at predicting those shifts. You can't
30:50just say oh well I expect it to do good
30:52on this task because it did well on this
30:54other task. explicitly validating the
30:57output is really um important and also
30:59provides a basis for choosing amongst
31:01many different measurement possibilities
31:03and arguing that our measurements are
31:05credible. And so just to briefly
31:08conclude, history plays an important
31:10role in determining the institutional
31:12the economic environment that conditions
31:14economic growth. The literature to date
31:17has largely focused on establishing that
31:19history matters and provides various
31:20insights into why. Moving forward,
31:23there's still a huge amount of data that
31:25remains unexploited that I think can
31:27really enrich our understanding. Um, and
31:30innovations in curating novel data at
31:32scale have the potential to unlock
31:34significant new insights into kind of
31:37rich patterns of persistence and
31:39transformation across time. Thank you.
31:51Great. So, I think I'll go. Um, let me
31:54see if I can share.
31:57Uh, I'm going to replace yours, Melissa.
31:59>> Awesome. Thank you.
32:00>> Yeah. And then this is it.
32:06Okay. Let me move that out of the way.
32:08Okay. So, everyone can see me and hear
32:10me, I hope. Um, so I'm Nathan Nun. I'm a
32:13professor at University of British
32:15Columbia. And so I'm very happy to be
32:17here. I I know there were some questions
32:19about using the slides. I'm happy if
32:21anyone wants to use these slides
32:23afterwards for their own teaching or if
32:25you want more detail on any particular
32:28topic, I you can email me and there's
32:30slides behind uh more extensive slides
32:33behind uh these slides. This is kind of
32:36summarizing
32:37uh maybe roughly half of what I would
32:40teach in one semester. just want to try
32:42and give you a bit of a a sense of what
32:44I think is interesting within this uh
32:46field. Okay. And let me put this on. So
32:50I um don't go over. So I think one of
32:53the things that lots of us are
32:54interested in is uh the variation that
32:58we see in this figure. So this is just
33:00from David Wild's economic growth
33:02textbook and it's just uh plotting per
33:05capita income and then the width of
33:07these bars is the number of people that
33:09live in that country. So China is quite
33:11quite wide. So if you thought of like
33:13we're going to line everyone up from uh
33:15who's the richest to the poorest. Uh and
33:19then the height is going to be the
33:20wealthiest. This is the kind of the
33:22parade of income. That's what he calls
33:24it that you'll see this group here is
33:26about 40 times uh wealthier than this
33:29group here. Okay. So this isn't
33:30surprising to anyone. I think one thing
33:32that's interesting might think about
33:34well where did this distribution where
33:36did this variation come from? Well, if
33:38we look back in time, it didn't always
33:40exist. Okay, so this is just some data
33:42from Angus Madison. This is real per
33:45capita GDP throughout time for different
33:48regions. It's hard to show it for every
33:50country. And you can see it's sometime
33:52after here. If we look at this on a log
33:55scale, it's a bit more clear. So that's
33:57the smaller changes um become more uh
34:01obvious. And it seems like something
34:03after 1500, right? So in 1500 you start
34:06to see differences 1600 and then
34:09particularly 1700. Okay. And so um
34:14recent research has really tried to
34:15understand this better. So this is a
34:17recent paper in the quarterly journal of
34:19economics. It tries to estimate
34:21productivity of basically this group but
34:24which is driven by England right and you
34:27see the same thing right? So it's using
34:29very granular historical archival data
34:32that it's something around this period.
34:34So a 1600. Okay. So what happened in the
34:38world around that time and this is why I
34:40think history is important to understand
34:42development today. Well, there's a few
34:44things. One is what's called the
34:45Columbian exchange. So this has been
34:47called by many historians the first or
34:50second most important invented human
34:52history. And that's Christopher Columbus
34:55sales to the Americas. And basically
34:57these two different regions of the world
34:59which were previously separated become
35:02integrated. So lots of the things that
35:04we associate with Europe, Africa, Asia
35:08are actually from the Americas. So you
35:11think of cacao for example, lots of it
35:13grown in Ghana or West Africa uh didn't
35:16exist indigenously uh there. Maize
35:20similarly, cassava similarly, right? Um
35:23and so other things like tomatoes, there
35:25weren't tomatoes in Italy prior to this
35:28event. Uh no spicy food. So in Asia, for
35:31example, you didn't have peppers. So you
35:34didn't have the same spice that you do
35:36today. Okay? So for example, Korean food
35:38had no spice at all prior to this uh
35:41period. Um and then disease. So there's
35:44disease and decimated an estimated
35:47upwards of 90% of the population of the
35:49Americas. Okay. And so this lead allowed
35:52then or dise the prevalence of disease
35:56provided a lot of kind of open land or
35:58empty land which was then settled by
36:01Europeans. Okay. So we know this what
36:03followed this was what's called the
36:04triangular trade. Okay. So where
36:08manufacturer goods were sent from Europe
36:11to Africa. They were then traded for
36:13human beings and enslaved people were
36:16then sent to the Americas. And the land
36:19here now has been depopulated because of
36:21the disease. So the enslaved people and
36:24then immigrants from Europe basically
36:27produced these raw materials here and
36:29then those were sent to Europe and then
36:31manufactured. So this is called the
36:33triangular trade and this occurred over
36:36about three four 400 year period. Okay.
36:39And so now as Melissa mentioned there's
36:41been a lot of research on this
36:42historical event. So I could tell you
36:45that oh this was super important. That's
36:47why we see this divergence here. But
36:50that's just me stating it. There's no
36:52evidence about that. Okay. Uh so a
36:55number of studies have started to look
36:56at this in more detail. Melissa
36:59mentioned her study within the Americas.
37:01Within the Americas, European contact or
37:04this episode had very different effects,
37:07right? In Peru and Bolivia where she
37:08studied it had certain effects. In the
37:10Americas, it had different effects. It
37:12depends on who you who you are as well.
37:14Are you now descended from formerly
37:16enslaved people or for from Europeans?
37:19Okay. I think what's been the most well
37:21documented is that the Atlantic trade,
37:23this three corner trade benefited
37:25Europe. And so, uh, Don Asamaglu, James
37:29Robinson, and Simon Johnson kind of
37:31studied this in 2005. This is part of
37:34what they won the Nobel Prize for,
37:35right? Just showing documenting the
37:38importance of Atlantic trade to these
37:41European countries. There's a more
37:42recent study that actually looks at
37:44genealogical data at the individual
37:47level. People who owned enslaved people
37:50in the Americas, then they invested
37:53those profits back in Europe and that
37:55led to economic growth. So this is a
37:57very recent paper that shows that uh
38:00work I have basically looks at different
38:02regions or countries or areas of Africa
38:05and documents that the regions or places
38:09that had more enslaved people taken
38:12within Africa are the poorest today.
38:15Okay. And so um and then there's a great
38:17paper if you're interested by Luigi
38:19Pascali that really tries to get
38:21identification by looking at the switch
38:24from sale to steam that had complex
38:27differential effects on different
38:29countries depending on who you're
38:30trading with and in a panel setting. So
38:33looking over time and space you can then
38:35estimate the causal effects of trade on
38:38all countries in the world. This is in
38:40the 19th century basically. And a very
38:43striking thing he finds very surprising
38:45to him for most countries in the world
38:48it was detrimental to growth. Right? To
38:51the richest countries or the countries
38:53with best best institutions like Europe
38:55it was beneficial but to lots of the
38:57other countries in the world uh
38:59countries that were colonized at the
39:01time for example or former colonies it
39:03was harmful to growth. Okay. So, so if
39:06you're interested in this, I could spend
39:08the whole time talking about this, but I
39:10think this is what the literature and
39:11what we've learned over the past uh 20
39:14years roughly. Okay. Um, but as Melissa
39:17mentioned, the interesting thing is not
39:19well that these past events or the
39:22interesting thing moving forward is not
39:24that these past events had effects today
39:25and have shaped the global uh economic
39:28landscape, but exactly how and why. So I
39:31think the the future research for all
39:33the grad students out there is thinking
39:35much more uh deeply about mechanisms and
39:38which mechanisms are universal or which
39:40mechanisms are location specific. Okay.
39:43And so I'm going to talk about just one
39:45line of research that I've been thinking
39:47about which is thinking about one way to
39:50think of think about all of this is well
39:52why are things so sticky? So I remember
39:54actually giving a presentation about my
39:56research in a conference that was um put
39:59on by the government in the Democratic
40:01Republic of Congo and there were a
40:02number of people that legitimately asked
40:05why are you talking about the slave
40:06trade that happened you know it ended
40:09100 over 100 years ago now right 150
40:12years ago that's in the past that can't
40:15affect us now right and Melissa talked
40:17about well traditional models in
40:19economics like the solo model have one
40:21equilibrium so if you shock it, do
40:24something bad, you move away from the
40:25equilibrium, but soon you'll move back
40:27to that equilibrium, right? And so, so
40:29it's a legitimate question. Why are
40:31these events that happened long ago, why
40:33do we see evidence that they still shape
40:36the economic uh landscape today? It's
40:39it's a very legitimate question. It's
40:41not obvious that that the empirical work
40:44would have found that. So, why are
40:46things so incremental and sticky? Well,
40:48I'm going to talk about the micro
40:49foundations. So let's go to like you
40:52know as micro as we can go we're moving
40:54away from states you know more granular
40:57more granular than regions all the way
40:59to people. So how is it that we make
41:02decisions? Okay. And so when we make
41:05decisions, well, we have to make
41:06decisions in a very very complex world,
41:09right? Uh it's estimated that we make
41:1230,000 decisions every day. Uh and lots
41:16of these are high stakes, not all of
41:18them. And almost all of them we don't
41:21have all the information that we need,
41:23right, to to make that this uncertainty.
41:25And so what's happened is we given
41:27cognitive limits, we've developed
41:28heristics or shortcuts that help us make
41:31decisions, right? We're not actually
41:33these rational calculating agents that
41:35we write down in our uh econ 101 models,
41:39right? That we actually um have other
41:42methods of making decisions. And so
41:45there are things like values, beliefs,
41:47norms. So in your society or in your
41:50area, what's the right thing to do? Um,
41:52you might have traditions that are
41:54passed down from your previous
41:56generation. So, what do your parents
41:57tell you or aunts and uncles uh tell you
42:00is the right thing to do, right? And so,
42:03you actually place a lot of value in
42:04what you're taught, right? Or you could
42:06have those um those who you look up to,
42:09teachers, priests, etc. could could tell
42:12you. Um, and then not only that, but
42:14we're also influenced by the behavior
42:16and values of those around us. So if
42:19there's someone who I really look up to,
42:21admire, I ask them what I should do in a
42:23particular situation, they'll give me
42:25advice. I actually uh put some value in
42:28that advice that has some influence on
42:29me. Okay. So all of that is what's
42:31called culture. It's basically uh
42:34values, beliefs, things that affect our
42:36behavior that's transmitted between
42:38people either from a previous generation
42:41to the current that's hor vertical or
42:44oblique or between people alive at the
42:47same point in time. That's horizontal.
42:48Okay. So, there's actually a consequence
42:51of this is that we tend to value the way
42:54things are in the world. If you see
42:56everyone doing something, you place
42:57value in that. If you see all your
42:59ancestors did something a certain way,
43:01you place value in that. And you uh
43:04listen to that advice. It it actually
43:06influences you. Okay. Um and so there
43:08isn't there is evidence that actually um
43:12we're hesitant to change because of
43:14that, right? If you see the way the
43:15world is, that seems natural. There's
43:17there's importance placed in that. Uh
43:20and so if you're interested in that, you
43:22can have a look at those those papers
43:24there. They kind of yeah neat evidence.
43:27So let me just give you an intuitive
43:29example. And so some of you may have
43:30already seen these if you've seen
43:31previous lectures, but I think they're
43:33the the kind of great examples of the
43:36value of culture and tradition. So why
43:38why aren't we just rational? That would
43:40be much easier. We just calculate
43:42things. Why do we even listen? If
43:44somebody tells us something, why do we
43:45listen? Why do we care what our customs
43:48in our society are? Why do we follow
43:50them? Uh one example from the Americas
43:54is uh this from this picture here. This
43:57is how maize is uh processed in many lat
44:01regions of Latin America. So maze will
44:04be dried. So corn will be dried. it'll
44:06be stored and then when you want to uh
44:09consume it then you'll boil it for 40
44:11minutes an hour in water and then it'll
44:14cook and then you'll mash it up into a
44:17dough called masa uh and and make you
44:21know different things out of it
44:22tortillas for example tamales um and but
44:26what is custom or tradition is to throw
44:30dirt or limestone in the water okay and
44:34so that's what you see here and you
44:36actually even in supermarkets today buy
44:38limestone that you throw into the water.
44:41And the reason is if you ask people it's
44:43like well that's just how we cook it. It
44:44tastes better etc. Trai- tra
44:46traditionally but uh it turned out
44:49there's a benefit to this. So if you eat
44:50a diet that's primarily maize. Uh it's
44:54deficient in niacin and if you don't
44:56have enough niacin you'll develop a
44:58disease called pelagra. And that disease
45:01basically will give you dementia, skin
45:04lesions and kill you. Okay. So, but if
45:07you put um alkali or have an alkali
45:10solution that you cook the maze in and
45:12you get that from limestone, then that
45:15helps the body absorb the small amounts
45:17of niacin that exist. Okay. So, those
45:20places that developed this tradition of
45:22cooking, those places basically didn't
45:24develop pelagra uh and then uh and they
45:28did better. Okay. So nobody knew why
45:30that was. Those traditions were just
45:32passed down from generation to
45:33generation. So remember in that
45:35Columbian exchange figure I showed you,
45:37we saw that corn traveled from the
45:38Americas to other parts of the world
45:41like lutu, northern Italy, southern
45:45United States. Interestingly the
45:47traditions did not transfer, only the
45:50food and in all those places people
45:52developed pelagra. It wasn't until
45:55actually 1930,
45:57so hundreds and hundreds of years later,
46:00that uh they realized what the disease
46:02was being caused from, right? And it
46:04took these scientists uh a huge amount
46:07of research to uh to determine that. So
46:10this is one example of the value of
46:12tradition, right? Uh so so to kind of
46:14figure out what's optimal to eat would
46:16take a a lot of calculation, but if you
46:19just rely on the customs that you have
46:21always had. So there's other examples
46:23like uh indigenous population northern
46:26Canada where they believe in this thing
46:28called scapula manscy where they uh want
46:32to decide where to hunt and what they do
46:34is throw caribou bones in the fire and
46:37the bones crack and then that provides
46:39them maps of where to hunt. Right? So
46:42why is that beneficial? Well, when
46:44you're hunting, you want to implement,
46:46you want to randomize because if there's
46:47any pattern in what you do, the animals
46:49will understand uh and they'll avoid
46:52you. So, this is a way of randomizing,
46:54right? Um and then there's a famous um
46:57religion called the Gamma Terta, which
46:59is a religion that believes in a water
47:03goddess. Water is holy and so they have
47:05certain rituals around planting. And it
47:08turns out that was really beneficial for
47:10controlling crops, right? And so that
47:12was found out in the 1970s when the
47:14Indonesian government made it illegal to
47:17have those um have those rituals um
47:21associated with planting in coordination
47:23with the World Bank and then suddenly
47:25the fields became overridden with pests.
47:27Okay. So so there's details in there I'm
47:29not going into just because of time that
47:31gives you some sense. Okay. So I just
47:33want to give you one example in a model
47:35maybe a bit more intuition. So this is a
47:38very simple model. It's an older model
47:40by Alan Rogers and in a paper with Paulo
47:42we look at that. So the model is
47:43basically this um each period there's
47:47you could be in one of two states and in
47:50in one state action A is better in
47:53another state action B is better. So all
47:54you have to do is choose action A or
47:56action B. Okay. So there's two
47:58strategies. One is you can invest uh
48:02kappa. You go out you do trial and error
48:04you learn and then you get the benefit
48:06beta. Okay. So that's one strategy. On
48:09the x-axis here is the proportion of
48:12people that that rely on culture, what
48:14we're calling tradition, which is they
48:16learn from the previous generation.
48:18Okay. So the so this is the payoff to
48:21people that go out and figure out things
48:23on their own. So it doesn't matter what
48:25others are doing. I've gone out figured
48:27out the best action and have
48:28implemented. I get B, but I bear the
48:31cost kappa. Okay. And so the one
48:34question is is there could there be an
48:36equilibrium where there's no culture in
48:39society where there's just rational uh
48:41economic agents? Well, let's look at the
48:44payoffs when that's the case. So what
48:47can happen here is if everyone [snorts]
48:50is going out and figuring out the best
48:52thing, then if there's one person that
48:54says, "Oh, I'm not going to do that, but
48:57I'm just going to copy or listen to copy
49:00what the previous generation did." then
49:03that individual gets beta and doesn't
49:06bear the cost kappa. Okay, they don't
49:09have to do that search cost. There is
49:10this wrinkle that the environment can
49:12change and if the environment changes
49:15then what the previous generation did
49:17and you're copying it isn't going to be
49:19beneficial. So as long as the
49:20environment is sufficiently stable that
49:23so for example if the environment never
49:25changed delta is zero this is always
49:27going to be higher right and then as as
49:30there's less people to copy then your
49:33payoff goes down and it turns out
49:35there's an interior equilibrium okay so
49:37but the key point here is if you rely on
49:41culture you can free ride on a whole
49:43bunch of information right from the
49:45previous generation and so that's kind
49:47of what this model does so yeah In other
49:49models, the fact that uh traditions
49:52exist means that they must be pretty
49:54good, right? So, so the fact that we had
49:57this and it persisted means it must have
50:00a benefit. Okay, so that's kind of the
50:02intuition of these types of models and
50:03how you can learn from others. Okay, the
50:06downside is well this generates
50:08stickiness, right? When the environment
50:10changes that you're not going to
50:12immediately switch to the new better
50:14action. And this is just showing you
50:16basically with different prevalence of
50:18tradition uh it's going to be slower.
50:21Okay, I'll come back to that in a
50:23second. Okay, so this has important
50:26consequences. Um and I'm going to talk
50:28about why this is interesting and um so
50:33one is this actually generates what's
50:36called a norm psychology. Okay. So in
50:39psychology, so that's actually we start
50:41to care if others one is if we're doing
50:45the same thing as others, right? You can
50:47see how culture is then going to
50:49generate that. We start to care if
50:51others are doing what the majority is
50:53doing. We start to care if people aren't
50:55following traditions, right? And this is
50:58all basically these all feed into what
51:00we call social norms, right? A belief in
51:02what's right and wrong. So in this
51:04society here, the belief was it's right
51:07to cook your food this way, right? Or
51:09cook maze this way. And so that norm and
51:12if someone didn't, well maybe they would
51:14be looked down upon, right? And so these
51:16norms which you've you know we studied
51:18are really uh uh founded in or are
51:22basically the flip side of of culture.
51:24Okay. Uh so culture and norms in turn
51:27shape institutions. Okay. So if you
51:29believe that certain ways of acting are
51:32are correct, then you might have a
51:34formal rule that says you're not allowed
51:37to do this, right? Just think of uh the
51:39norm that you shouldn't harm others, you
51:41shouldn't murder others. Well, then
51:42we're going to implement laws that
51:43enforce that. Okay? And then but then
51:46those institutions will affect the cost
51:49and benefit of different beliefs. So
51:51then they can affect cultural traits,
51:53right? How you think, what you think is
51:55right and wrong. So those co-evolve.
51:57Okay. Um and then there's dynamic
51:59implications. So one is cultural
52:01mismatch. So in these cases here the
52:04actions that people are taking because
52:06they're relying on culture uh will be
52:10misaligned with the environment when the
52:12environment changes. Okay. But also
52:15that's in that's in a dynamic sense also
52:17in the cross-section if there's policies
52:20for example the World Bank is going into
52:22one location versus another location
52:24those policies might be mismatched for
52:27the local context for the local culture
52:29for the local environment. Okay. Um, and
52:33then the other thing is, and I think
52:34we're still starting to understand this,
52:36Leo Burstein has basically documented
52:39and co-authors a number of papers that
52:41our perceptions of the norms held by
52:44held by others or our perceptions about
52:47what others think is right and wrong are
52:49often incorrect. So, how could that be?
52:53Well, there could be a mismatch in an
52:55environment like this. So he shows in
52:56Saudi Arabia that basically m men think
53:00that other men will look down upon them
53:02if their wife works. But that's not
53:05true. So how could that be? Well, maybe
53:07the environment is changing, right?
53:10Maybe now it's an environment such that
53:13uh people's values are changing and then
53:15gender equality is becoming more
53:17important in Saudi Arabia or more
53:19prevalent. Okay. Okay. So I'm just going
53:22to talk quickly about how the interplay
53:25about culture and institutions and give
53:27you some examples of what this means
53:29concretely. Okay. So here is the effect
53:32of culture on institutions. There's
53:35actually a lot of work. So um well-known
53:38uh economic historian Aarfe has studied
53:41this amongst the murby and Genoies
53:44traders in the Mediterranean in the
53:46medieval period. I think one of the best
53:48examples is actually from the United
53:50States. So there's a great book by David
53:51Hackett Fischer Albbnc where he looks at
53:54the first settlers in the US. So there
53:56was actually four waves. There was
53:59Puritans, Quakers, uh Anglican Anglican
54:02Cavaliers he calls them and then the
54:04Scotch-Irish. And so they had very
54:06different values and beliefs about
54:08what's right or wrong. And so the
54:10Puritans believed in order and strong
54:13government and because they were
54:15Protestants that everyone should read
54:17the Bible. Okay? So they settled in
54:19what's today Massachusetts and New
54:21England. And there even today you have a
54:24fair amount of of government
54:26involvement. Taxes are higher and
54:28education is very very very strong,
54:31right? And so there's other pe other
54:33settlers like the Quakers. They
54:35basically just wanted freedom from the
54:37government, right? And so they settled
54:40in Pennsylvania. Uh and so you have very
54:42different institutions that were set up
54:44there, right? and uh and similar the
54:46ScotchIrish in the Appalachian region,
54:49they really just wanted um to police
54:52things themselves. So, it's more of a
54:54culture of violence or a culture of
54:56honor, it's called. Okay. And so, he
54:58really documents how these how these
54:59persist and how the it's kind of a
55:02natural experiment. People come over and
55:04all they bring with them are their
55:05values and beliefs and you know, things
55:07embodied in their in their minds and
55:09they're setting up what they think is
55:11the ideal society. So you really see how
55:14uh the institutions are endogenous to
55:16this. Okay. I have some work on zero sum
55:19thinking and political preferences. Uh
55:21and so basically if you think the world
55:24is zero sum that means those
55:26billionaires got their money by taking
55:28from others then you're going to have
55:30different views about taxation of the
55:32rich right then you're believe it's more
55:34acceptable to tax the rich. Okay. Uh and
55:37so we trace that. But also if you have a
55:39zero sum view of the world, you might
55:41think immigrants the benefits they get
55:44after coming to the US come from others.
55:46So then you'll have certain views about
55:48immigration policy. Okay. And so yeah,
55:51so this paper basically basically um
55:54tackles that. Uh so what h does it
55:57matter if policies or institutions are
55:59not uh in line with the underlying
56:01culture? So now we're thinking about
56:03development policies and there's a great
56:05forthcoming handbook chapter or in the
56:08handbook of culture and economic
56:09behavior that I'm editing with Beni and
56:13uh Paulo Giliano. Basically, they do one
56:16neat exercise where they look at the um
56:20director of World Bank Projects and they
56:22have I can't remember if they use a
56:24different name, but essentially the the
56:25lead of World Bank projects. They look
56:27at their name and they look at the
56:29distance of their name from the names in
56:34the country in which the project's being
56:36implemented. Right? So what's the kind
56:38of cultural distance of the person
56:40heading the project to the population
56:42that's receiving the project and the low
56:46the the closer the distance or the more
56:49yeah the closer the distance in other
56:51words the more the individual probably
56:54uh knows about the culture of the
56:55location the better the project performs
56:58and the World Bank has these rankings A
57:00B C D right so you actually can can uh
57:03have an objective measure at least that
57:06they uh that that uh that they put on
57:10the project. Okay. So, so the more
57:12aligned the director is with the culture
57:14of the location uh the more successful
57:17these are. Um and then if you're
57:19interested on the reading list is this
57:21fish and trees uh handbook chapter. If
57:24you want to know what the heck does that
57:25title mean, uh you can read that. It's
57:28with Jacob Mosonea and James Robinson.
57:31And one way to think about this is well
57:33when there's a misalignment and this is
57:35language from Albert Hirschman. Well you
57:37can think about well what do we change?
57:39We could change the policy or we can
57:41change the culture right of the location
57:44and one's called trait making one's
57:46called trait taking. So trait making
57:48would be we need to change the culture.
57:50The the uh so yeah trait m trait making
57:54would be we need to take change the
57:56culture. Trait taking is well we have to
57:57take the culture is given and that's
57:59important and then let's change the
58:01policy okay and I think you can think
58:04about the research that's being done
58:06often without thinking about it we are
58:09in this trait making world there's a lot
58:11of RCTs which are trying to change norms
58:14provide information to change behavior
58:16etc and that's saying well you know uh
58:19if there if there's a misalignment or
58:21something we don't like let's just
58:23change the local culture okay and so in
58:25that handbook chapter we try and think
58:27about that a little bit. Okay. So, uh
58:31the interplay of culture and
58:32institutions well uh institutions will
58:35affect culture by affecting the relative
58:38payoffs of one value or not. And this
58:41actually leads to a lot of really
58:42counterintuitive stuff actually. So, uh
58:45so some intuitive stuff is like Natalie
58:47Ba's work where government pensions come
58:49in and then norms of children looking
58:52after their parents weaken because they
58:54don't need to, right? And so she looks
58:55at this in Indonesia and Ghana for
58:57example, but there's others like uh
59:00Melissa mentioned this paper by Sarah
59:02Los and Eduardo Montero. Congo Free
59:04Straight Free State. They're extracting
59:06rubber through a lot of force forced uh
59:11uh extraction, coercion, violence. They
59:13killed uh 10 million people. And what
59:16they found is this led to more
59:18cooperative behavior. Okay, so this led
59:21to more pro-social values. It was bad
59:23for economic growth. Uh but it led to
59:26and the reason was well the returns to
59:27cooperating basically went up. So what
59:31would happen is um if any individual
59:34didn't collect enough rubber basically
59:36they would be killed or their hands
59:37would be cut off. So now there's
59:39benefits to helping your neighbor
59:41because the cost is so high of them not
59:43achieving their goal. Okay. Um okay. And
59:47then there's formal work by Guido
59:48Tabellini basically looking at at this
59:51in a mathematical model right and so
59:53basically there's this you know an
59:56example of this is good institutions can
59:58crowd out good culture if the good
1:00:00institutions ensure good behavior right
1:00:03so if the institutions are ensuring good
1:00:05behavior you don't need to inculcate
1:00:07these norms or these values or parents
1:00:09don't need to right and so it's this
1:00:11weird kind of counterintuitive thing in
1:00:14economics we often think all good things
1:00:15should go together. But actually, uh
1:00:18it's it's it's not obvious that that's
1:00:21going to be the case. Okay. One way I
1:00:22think of it, there's two ways to get uh
1:00:24a good to a good equilibrium. Top down
1:00:26enforcement or bottom up intrinsic
1:00:29motivation. And those two things can be
1:00:31uh substitutes. Okay. And the last thing
1:00:34I'm going to talk about then I'll end is
1:00:36what determines what traits thrive. So I
1:00:38kind of implicitly being say been saying
1:00:41well cultural traits if they're
1:00:42beneficial are going to become more
1:00:44successful. That's why alkali processing
1:00:46a maze, throwing limestone in the in the
1:00:49in the water uh persisted all all this
1:00:53time. If it was detrimental, it probably
1:00:55wouldn't persist. But when we think
1:00:57about well, if traits do better, they're
1:00:59more likely to survive. Uh are we
1:01:02talking about individuals or are we
1:01:04talking about groups? Okay, so some lots
1:01:06of things. Think of like pro-social
1:01:08norms, helping others or cooperation in
1:01:11a prisoner's dilemma. That's not an
1:01:13individual equilibria. That's not an
1:01:14equilibria. It's not good for
1:01:16individuals. But if as a society we all
1:01:18cooperate in a prison dilemma or a
1:01:20public goods game, the society will be
1:01:22better off. Okay. So in the literature
1:01:25and this really hasn't been studied much
1:01:26at all in economics. Uh the what what's
1:01:30going to kind of dominate depends on how
1:01:32much group level selection there is. So
1:01:34let me give you an example. Think about
1:01:36a co cooperation in a presidential
1:01:38dilemma. So people meet and then they
1:01:41can cooperate or not, right? And if the
1:01:44setting looks just like this, two people
1:01:46meet then they cooperate and if one
1:01:48person does better then um or not
1:01:52they'll cooperate or not. If one person
1:01:53does better then that trade will or
1:01:55spread then we know that defect defect
1:01:58is kind of going to give you the highest
1:02:00payoffs. But what if the setting looks
1:02:02like this that people live in
1:02:04communities there's a norm psychology so
1:02:07everyone kind of uh converges on the
1:02:10same type of norm like in this community
1:02:12we're nice to each other we cooperate
1:02:14other communities maybe they're mean to
1:02:16each other they don't cooperate and then
1:02:18there's competition between communities.
1:02:21So what's going to arise in this
1:02:22environment is actually uh cooperation
1:02:25is going to do better. That's a behavior
1:02:28or a trait that's not individually
1:02:30optimal, but it's optimal at the group
1:02:32level, right? So, if everyone
1:02:33cooperates, the group does better, but
1:02:36anyone individual if they don't
1:02:37cooperate, uh, they'll do they'll do
1:02:40better. Okay? And so, this is another
1:02:42kind of, you know, consequence of
1:02:44culture. Remember, I said culture leads
1:02:46to norm psychology and to groups forming
1:02:49and groups having similar norms. uh so
1:02:52culture will actually lead to an
1:02:54environment like this which then can
1:02:56allow pro-social traits to arise. Okay.
1:02:59So it's this kind of interesting thing
1:03:01and then one thing you see actually or
1:03:04implication of this is one form of
1:03:06competition particularly historically
1:03:08between cultures or societies is warfare
1:03:11actually. So this is kind of like with
1:03:13warfare either our whole whole society
1:03:15rises or falls together, right? Uh and
1:03:18so this has this implication that war
1:03:20might lead to pro-social traits actually
1:03:23and there's evidence of this. So kind of
1:03:25a case study with Ral Sanchez with
1:03:27Sierra and I in Bukavu in Eastern DRC.
1:03:31There's some metaanalysis evidence and
1:03:33then evidence from a lab experiment,
1:03:35right? And so so I'll just show you the
1:03:36lab experiment that I'm ending is this
1:03:39is Patrick Francois uh teams uh
1:03:44uh play this experiment. They the teams
1:03:47are are basically grouped together and
1:03:49then they play a public goods game,
1:03:51right? So you can contribute and then
1:03:53that amount is increased and then shared
1:03:55equally. That's basically a prisoner's
1:03:57dilemma. And so the equilibrium is for
1:04:00no contribution and that's what you see
1:04:02here, right? But what if you you tweak
1:04:05it and say okay what's going to happen
1:04:07is you'll be formed into a teams you'll
1:04:10contribute then you'll get a payoff but
1:04:13then if your payoff is not equal to or
1:04:16higher the payoff of another team or the
1:04:19total contribution is not equal to or
1:04:21higher the total contribution of another
1:04:23team then you'll get zero. So suddenly
1:04:25you've introduced a comparison between
1:04:27these teams. uh and what you find here
1:04:31when there's intergroup competition the
1:04:33contribution levels in the public goods
1:04:36game is now suddenly much higher right
1:04:38and so they have a bunch of analysis the
1:04:40dynamics matter etc etc but this is just
1:04:43showing you this intergroup competition
1:04:45effect right so not only depends what
1:04:48how well I do within my group but now I
1:04:52care about the well-being of the group
1:04:53as a whole because if the well-being of
1:04:56or the payoffs of my group as a whole
1:04:57aren't high enough we're all going to
1:04:59get zero. Right? So you can kind of see
1:05:01the analogy with conflict. Okay.
1:05:04Okay. So those are I just want to give
1:05:05you a a bit of a sampling. So basically
1:05:08just to review what's happened is uh
1:05:11started with current distribution of
1:05:12income. I think it's fairly clear by now
1:05:15that this really is uh rooted in events
1:05:19historical events that h started in
1:05:211500. Um and if we think about
1:05:26um exact mechanisms, I think one useful
1:05:30line of research that's what I focused
1:05:31on is kind of thinking about individual
1:05:33decision making, individual culture, how
1:05:35we make decisions, how our decision-
1:05:38making leads to norms, how those norms,
1:05:41which you many people call informal
1:05:43institutions lead to formal
1:05:44institutions, how that feeds back and
1:05:47and why that's important for policy.
1:05:49Right? So I think all of this is uh is
1:05:53kind of open area for for future
1:05:56research and all the readings I put on
1:05:58are kind of like summaries that just
1:06:00think that think about these issues and
1:06:03hopefully um stimulate your own
1:06:05research. So I'll stop there. Thank you.
1:06:12>> Thanks so much Nathan um and Melissa.
1:06:15We're gonna kind of transition now to
1:06:17the Q&A portion. And so like last week,
1:06:21we'll have some people who are kind of
1:06:23live in the classroom that are able to
1:06:25ask questions themselves. And then I'll
1:06:27try to collect some of the questions
1:06:28from the chat as well.
1:06:30>> Yeah,
1:06:30>> you guys have had the opportunity to to
1:06:33read. So if anything crops up that
1:06:34you're interested in, please flag it.
1:06:36>> Um Gabrielle, we'll start with you. And
1:06:39if we could just try to keep these
1:06:40brief, that'd be great.
1:06:43>> Hi. Thank you so much. Um my question
1:06:45was about historical events and the kind
1:06:47of the stickiness and the concept of
1:06:49stickiness. Um
1:06:52is there a measure of how historical
1:06:55events affect over time? Because um I
1:06:59was thinking about for example the MIA
1:07:00paper, how we see like the effects
1:07:02today, but do we see whether these uh
1:07:06historical events kind of attenuate over
1:07:08time? Like is there a point where we can
1:07:09not blame the Spaniards um or we blame
1:07:13them more over time? Um
1:07:17>> yeah I mean I think it really depend how
1:07:19much you can look across time will
1:07:20really depend on the context and what
1:07:22variables are available. So for instance
1:07:24I don't have high quality consumption
1:07:26surveys right before 2000 um but I do
1:07:30bring in various data that we do have
1:07:32over time like on education on land
1:07:34ownership and so it's like imperfect
1:07:37because we can't necessarily measure the
1:07:39same thing in the same way across time
1:07:42but I think the goal is always to um you
1:07:46know provide as rich of information as
1:07:50possible um and give a nuance story,
1:07:53right? And so it's not just about this
1:07:55thing the Spanish did, but how that
1:07:57interacted kind of with the world going
1:08:00forward.
1:08:02>> Yeah. I'll just say really quickly,
1:08:04yeah, scholars have tried to do this in
1:08:06different settings and like Melissa
1:08:08said, you'll get different dynamics. So
1:08:10one is there's a shock and it makes you
1:08:13worse off, but slowly you're going to
1:08:15converge to the to, you know, you're
1:08:17going to overcome it. But the other
1:08:19another dynamic is and then so then if
1:08:21you looked over time the effects would
1:08:22get smaller and smaller. Another dynamic
1:08:24is the shock moves you to a new
1:08:26equilibrium. So the shock at first
1:08:28there's small differences but things get
1:08:30worse and worse and worse. So there the
1:08:32effects get uh worse and worse over
1:08:34time. Yeah. So in some of the stuff like
1:08:35the slave trade what I found is maybe
1:08:38not surprisingly
1:08:39um postc colonial you know in the
1:08:42postcolonial period the effects start to
1:08:44show up of the slave trade and then they
1:08:46grow over time and then once natural
1:08:48resources are discovered so most
1:08:51recently then they start to you know
1:08:53attenuate slightly and so yeah so yeah
1:08:56so there's kind of will be different
1:08:57dynamics depending on exactly what's
1:08:59going on. Yeah.
1:09:02Thank you.
1:09:04>> Thank you. Um I think there was a
1:09:06related question in the chat from uh
1:09:09Anendita
1:09:11who was also kind of asking about the
1:09:14puzzle with persistence. Um they said
1:09:17that you know countries obviously had
1:09:19institutions and social structures
1:09:21before colonization
1:09:23and so why have colonial institutions
1:09:26proved so persistent in many places
1:09:28rather than pre-colonial institutions
1:09:31reasserting themselves after
1:09:32independence and what determines
1:09:36um which historical layer persists.
1:09:39I mean I think we have you know there's
1:09:41lots of work on pre-colonial
1:09:42institutions as well and uh showing uh
1:09:45that that those uh matter too. So like
1:09:49in Africa uh Michelle and Pompaneo have
1:09:53you know a paper arguing that the
1:09:55pre-colonial institutions are actually
1:09:56more important than the colonial
1:09:58institutions at least in kind of their
1:09:59particular setting. um you know I have
1:10:02work um on uh Vietnam uh showing the
1:10:05importance of pre-colonial institutions
1:10:08often the place where these were the
1:10:10more the most I think persistent is
1:10:12places where the colonial state did not
1:10:13penetrate very well so yes the colonial
1:10:15the French colonial state colonized
1:10:17Vietnam they had limited ability to
1:10:19penetrate kind of in uh to kind of the
1:10:22countryside and so you see a lot of
1:10:24persistence in part because the colonial
1:10:26state wasn't overthrowing everything in
1:10:28kind of other places colonialism may
1:10:30have marked kind of a more extreme
1:10:32rupture. And so again, I think this is
1:10:33going to um to really depend on the
1:10:37context. And you know, there's been a
1:10:39focus on colonialism, but the
1:10:40pre-colonial institutions can matter a
1:10:42lot too.
1:10:44>> Yeah. I'll just add one thing. I think
1:10:46there's a tendency if you see one paper
1:10:48showing that colonial institutions
1:10:50matter, it doesn't mean that it's
1:10:52showing you that pre-colonial
1:10:54institutions don't matter, right? And so
1:10:56and then typically explanatory power is
1:10:58like I don't know less than 5% for sure.
1:11:00Uh and so there's a lot of other things.
1:11:03There's not any one thing that explains
1:11:05even the vast majority or even a big
1:11:07chunk right and so yeah and so I don't
1:11:10think anyone's done a decomposition
1:11:11about how important one or there is and
1:11:13whether that would even be possible. It
1:11:15would vary a lot but um yeah but I would
1:11:18say I agree with Melissa.
1:11:20>> Yeah. Perfect.
1:11:22Um, okay. Tom, uh, would you like to ask
1:11:24your question
1:11:28and you're on mute?
1:11:29>> Yes.
1:11:29>> Okay. Yeah.
1:11:30>> Great. Yeah. Thank you so much for the
1:11:32great lecture. Um, one small question I
1:11:34guess related to both the papers that
1:11:37were on the reading list and then what
1:11:38you talked about. I feel like culture is
1:11:41persistent is persistent because of kind
1:11:43of two reasons. One is the information
1:11:45is costly to acquire which is what you
1:11:47mentioned and then the other one is I
1:11:49guess more from a game theoretic
1:11:51perspective um any deviation is going to
1:11:53be punished so then people have an
1:11:55incentive to keep doing what they've
1:11:56always been doing otherwise they're
1:11:57going to be punished like how would you
1:11:59this two enforcement mechanisms or like
1:12:01this two explanations for the
1:12:03persistence
1:12:04how would you link them would you say
1:12:06that maybe one is culture and the other
1:12:08one is institutions or like how would
1:12:10you differentiate them
1:12:13>> uh well Yeah. So I yeah I can the annual
1:12:16reviews paper probably does that best
1:12:19but um basically to some extent I think
1:12:22from to have norms you need culture like
1:12:25why would you care what others think of
1:12:27you or you know and then you might think
1:12:30well they're going to punish you but why
1:12:32would it's not in their incentive to
1:12:34punish you in a purely rational it's
1:12:36they're punishing you because they think
1:12:37you're doing something wrong and so so
1:12:40you kind of need these values and
1:12:42beliefs that we think are important
1:12:44uh and if you call that culture or norms
1:12:46that's a yeah that's that's that's an
1:12:48open question but I think there is
1:12:49something exactly as you said there's
1:12:52some examples of culture where it's
1:12:53traditions I just like do something
1:12:56because others are doing it or but then
1:12:58there are these equilibria where we all
1:13:00think this is super important and we're
1:13:02going to punish others who you know uh
1:13:04who or think badly about others who who
1:13:07uh don't do that and that's more along
1:13:09in the lines of the social norms right
1:13:12and So um yeah so I think um in econ
1:13:17there's not much of a distinction made
1:13:18but I think if in the annual reviews
1:13:20paper we try and really uh make a
1:13:22distinction. There's also a paper in the
1:13:24annual re reviews of psychology that I
1:13:26co-authored with Michelle Gelfand which
1:13:28focuses on the norms specifically. Yeah.
1:13:30So in psychology there's a lot of work
1:13:32done on norm psychology and that's kind
1:13:34of like what aspects of our psychology
1:13:36or cognition prime us to buy into this
1:13:40idea of norms either following them or
1:13:42enforcing them. So yeah I don't know
1:13:44that kind of indirectly answers but
1:13:47yeah
1:13:49>> thanks um Chiny would you like to ask
1:13:52yours?
1:13:53>> Yes. Okay thank you coast. Uh thank you
1:13:56host. Um I'm Chinzy and I'm currently a
1:14:00second year PhD student student at the
1:14:02University of Melbourne. So uh I just
1:14:04have a question about the measurement of
1:14:05trust. So like I checked like the uh
1:14:08literature in behavioral economics and
1:14:11uh it seems that like we're adopting
1:14:14like different systems of like trust
1:14:16measurements. So like the first system
1:14:18is about like using a trust game or
1:14:20something to measure like how people uh
1:14:23are willing to uh invest in like a risky
1:14:27scenario. The other thing is about like
1:14:29uh based on survey or something. So uh
1:14:32how do we reconcile at least uh two
1:14:34different measures? Yeah. And and if
1:14:36anything um is that possible for us to
1:14:39um construct some like trust measures
1:14:42which are not restricted to our
1:14:44contemporary time based on historical
1:14:47archives or something
1:14:53go first and then Melissa [laughter]
1:14:55>> N has more experience with trust. You
1:14:58probably will have thought about the
1:14:59trust measures on on the one like I
1:15:02think it's like the trust games versus
1:15:04trust questions. They both just have
1:15:06pros and cons. Like the trust game you
1:15:09might think, well that's equal to
1:15:10everyone, but it's completely
1:15:11artificial. And then the trust question
1:15:14is kind of nice because that's about the
1:15:15real world. Like in general, do you
1:15:17trust people or do you trust your
1:15:18family? But then it's not specifying
1:15:20trust them to do what? So it's a little
1:15:22less precise. So there's a paper by null
1:15:24Johnson J O H N S O N that basically
1:15:28asks do those two things correlate? So
1:15:30if someone answers a trust question and
1:15:32and he finds they actually they actually
1:15:34do correlate. I think it's published in
1:15:36a psychology journal. Um so yeah so I
1:15:39don't think of like one's right, one's
1:15:40wrong. It's just like uh two Yeah. two
1:15:43different measures and uh they have pros
1:15:46and cons. Yeah.
1:15:49>> Yep. I think that's a a great summary. I
1:15:51mean it goes back to specifying what you
1:15:53want to measure and
1:15:55>> you know ideally we show robustness. Um
1:15:58but yeah I think that's it's great.
1:16:01>> Yeah.
1:16:03And then I I don't know if Melissa has
1:16:05thoughts but like I think it's a
1:16:07fascinating thing to think about looking
1:16:08at text and then seeing whether we can
1:16:11see the equilibrium amount of trust uh
1:16:14or yeah in a society is that reflected
1:16:17in text.
1:16:19I mean I think for that you have to have
1:16:20an operational [clears throat]
1:16:21definition that is observable right if
1:16:23we think of trust it's something in
1:16:25people's heads it's latent you have to
1:16:27take a stand on what exactly that means
1:16:30and text I think if you just still leave
1:16:31it as a latent concept and you tell the
1:16:34LLM to do it I mean who knows what it's
1:16:36doing you know we did this with economic
1:16:38policy uncertainty and it even goes it
1:16:40finds the famous Baker Bloom and Davis
1:16:42you know but every model comes up with a
1:16:44different thing and they're not that
1:16:45correlated right so I think it is
1:16:47promising to measure it in text, but you
1:16:49have to be willing to take an explicit
1:16:53stand on what that means where kind of
1:16:55two reasonable humans would see a text
1:16:57and agree whether it or not it aligns
1:16:59with your definition. If you leave it as
1:17:01just latent and you know you're not
1:17:03really precisely defining what it means
1:17:05and the context of a specific text, then
1:17:08it's hard to know like what we're
1:17:10measuring.
1:17:15>> Thank you. Um
1:17:18there was there were a couple questions
1:17:20in the chat about kind of AI and um
1:17:24biases of the data.
1:17:27>> Um there were a few people asked
1:17:29something to the effect of of what
1:17:31should best practice look like for kind
1:17:35of understanding detecting those biases
1:17:38and validating AI generated measures
1:17:40before we use them for economic
1:17:43inference.
1:17:44>> Yeah. I mean, so I think that, you know,
1:17:47what I would argue is that it's
1:17:51important to have a validation set where
1:17:54you have clear interpretable criteria.
1:17:58Um, you know, it could be using a
1:17:59measurement instrument. It could be, you
1:18:01know, that they're implemented by a
1:18:03human who's going and looking at the
1:18:04data and kind of recording, you know, is
1:18:06this about trust or not. uh but it's
1:18:08important to have those clear you know
1:18:11auditable validation sample because with
1:18:14the neural network we're not going to
1:18:15open it up and understand what it's
1:18:17doing and so to kind of to see this
1:18:19perspective I think you know one of the
1:18:20papers that I sent for the reading list
1:18:22goes through and discusses kind of very
1:18:25explicitly what validation means in the
1:18:28context of AI and why it's important to
1:18:30valid inference so that paper has more
1:18:32guidance also I would have put it in the
1:18:34chat but I don't think I'm authorized
1:18:36but the MBR our summer methods lectures
1:18:39this summer we talk about exactly the
1:18:41same question so validation with AI
1:18:44generated measurements so for people who
1:18:45are interested there's a lot more
1:18:47details there
1:18:51>> um Melissa do you have a um website or
1:18:55or something for the NBER um if you want
1:18:59to share that with me I put that in the
1:19:00chat there were a few questions
1:19:01>> I'll send that to you now
1:19:04>> um Jorge do you want to ask your
1:19:08Hello. Yeah. Yes, of course. I'm here
1:19:10from Colombia. Currently doing a predoc
1:19:13in the economic department at Harvard,
1:19:17we we start talking about like country
1:19:20level or and country data. And the more
1:19:23we advance in the in the literature and
1:19:26and the cases, we were going like more
1:19:30to context specific cities or regions or
1:19:35specific cases. So basically my question
1:19:37is are we
1:19:40cursed or condemned at country level and
1:19:44how we can resolve this tradeoff between
1:19:48the receip context institutions like
1:19:53that you are saying trade making or or
1:19:56trade taking and the external valid
1:19:59validity of the findings how we can
1:20:02solve this tension
1:20:09Um yeah so I think I think I think that
1:20:11brings a great point
1:20:14about uh what level do you want the
1:20:17analysis to be at? So, so like think of
1:20:20the um US south for example. If you look
1:20:25at like within the south and across
1:20:27regions, what you'll find actually I've
1:20:29looked at this a positive relationship
1:20:31between the prevalence of slavery
1:20:33historically and then economic
1:20:35development subsequently and then today.
1:20:38Uh and so but then we think as a whole
1:20:41that region that slavery or enslavement
1:20:44was not beneficial. Right. So I think
1:20:46you can see that a lot and I think it's
1:20:48something that Melissa and I have both
1:20:50thought about in our own studies that
1:20:51look at what one region and then and so
1:20:54another example is actually like um
1:20:57within Africa where where railways are
1:21:00or where any colonial infrastructure is
1:21:02those places are richer and I wouldn't
1:21:04take that as being colonialism was good
1:21:07necessarily it could be but what it
1:21:10probably does is relocate people and
1:21:12activity within the within the country.
1:21:15So kind of a more microlevel analysis.
1:21:17That's what you're always worried about
1:21:18is the effects you're getting is just
1:21:20like a relocation of of activity. Um and
1:21:24then but then you could moving at the
1:21:26country level allows you to compare
1:21:28general equilibrium general equilibria
1:21:30or like different equilibria to one
1:21:32another. But we have a very limited
1:21:34number of observations in that case
1:21:36right and we've kind of done a lot of
1:21:38cross-country stuff already. And so so
1:21:40probably like I think fruitful research
1:21:43is thinking about mechanisms
1:21:45theoretically and then you can better
1:21:48understand the microlevel uh data and
1:21:51and uh and better kind of uh combine
1:21:55that with predictions that you should
1:21:57see with theory and it's going to be
1:21:58non-obvious predictions. Once you write
1:22:00down a theory there's going to be some
1:22:01things you didn't see and I think that's
1:22:03going to be a way to better understand
1:22:06uh kind of these issues. Right. Yeah. So
1:22:09>> yeah and I mean I think that one thing
1:22:11that we'll see more of moving forward is
1:22:14people who have more granular data
1:22:16>> but across you know many different
1:22:18regions um or countries and
1:22:20traditionally that's been difficult
1:22:22because just the cost of creating this
1:22:24data is so high you know but as the
1:22:26costs fall and maybe we bring in more
1:22:28novel types of data you always have to
1:22:30be cautious because data might have been
1:22:32collected in different ways and have
1:22:33different meanings across space but like
1:22:36I'd say in the past five years we're
1:22:38seeing more and more studies that have,
1:22:40you know, multiple countries, multiple
1:22:42regions, um, but they have kind of more
1:22:44disagregated information like on, you
1:22:47know, on firms, uh, or on local
1:22:49governments, etc. So, I think that we'll
1:22:52see more of that. I mean, I think at the
1:22:53end of the day, yes, there's a huge
1:22:55identification problem with
1:22:57cross-country studies, but at the same
1:22:58time, if you look at like what's most
1:23:00cited, you know, it's always the
1:23:02cross-country studies. I'm sure that's
1:23:04probably true for Nathan's work, for my
1:23:06work. People say you can't you can't
1:23:07write a cross-country study. It won't be
1:23:09credible. But then at the end of the
1:23:10day, you know, people like the big
1:23:11picture questions and as Nathan said,
1:23:13kind of for many reasons, that's hard to
1:23:15do when you're focused on a narrow
1:23:16region. Even if the identification is
1:23:19more credible, it answers a different
1:23:20question.
1:23:24>> Thank you.
1:23:26>> Thank you. Um, another one from the chat
1:23:30was from Vuin who raised a kind of
1:23:35particularly interesting mechanism. Uh,
1:23:37they said that if if development itself
1:23:40affects beliefs about whether economic
1:23:42activity is zero sum, could we get a
1:23:45kind of self-reinforcing trap where zero
1:23:49beliefs reduce effort or cooperation
1:23:51which depresses development which then
1:23:53reinforces those beliefs? And if so,
1:23:57what could break that cycle?
1:24:00>> Uh yeah, so that's super smart question.
1:24:03So we actually uh I have a paper with
1:24:06Austinine Berseron and a bunch of
1:24:08co-authors Joe Henrik etc. where we look
1:24:11at the dynamics of zero sum with a model
1:24:14and then we actually have a spin-off
1:24:15paper which is exactly that that like in
1:24:18a zero sum world you'll have beliefs
1:24:20that demotivate you uh and then that
1:24:22could lead to less innovation and make
1:24:24the world less zero sum and so uh so
1:24:27anyways um so so if you're interested in
1:24:30that it's on my web page it's in the uh
1:24:34an econometric society world congress
1:24:36volume basically uh I think it's I think
1:24:39it's like literally what the question
1:24:40asks. And so if you're kind of
1:24:42interested in that, I think like one
1:24:43thing that could break you out of that
1:24:45is a temporary uh if the environment is
1:24:50temporarily nonzero sum, right? And so
1:24:53then when it's not zero sum, you won't
1:24:56have these demotivating beliefs that
1:24:57reduce effort and then you can have
1:24:59innovation and that moves you into a
1:25:01world that is uh more nonzero sum
1:25:05basically. And so and so we argue that
1:25:08could have been what happened with
1:25:09Europe actually uh after this uh age of
1:25:13exploration if you want to call it that
1:25:15is from Europe's point of view the world
1:25:17was no longer zero sum like it they had
1:25:19tons of land they were engaged in the
1:25:22transport of humans there's profits for
1:25:24a lot of people and from Europe's point
1:25:26of view the world was not zero sum and
1:25:27then it suddenly uh these demotivating
1:25:30beliefs like pursuing profit is is kind
1:25:33of a lowly thing and you shouldn't do
1:25:36that or being entrepreneurial was not
1:25:38good. Um, suddenly those values changed
1:25:41and so Dearra McCloskkey calls these
1:25:43bourgeoa values, right? And so and then
1:25:45that she argues led to the industrial
1:25:47revolution. So yeah, so yeah, so I think
1:25:50that's exactly very plausible that
1:25:52question. Yeah. Or the dynamics from
1:25:55that question.
1:25:59>> Great. Well, we are just about out of
1:26:02time. Sorry, Franca and the other
1:26:05hundred questions that were uh
1:26:07submitted. I wish we had more time. Um
1:26:10Nathan, Melissa, thanks so much for for
1:26:13coming today. Are there any um last bits
1:26:17of advice or or um anything that you'd
1:26:20want to share with the group before we
1:26:21go?
1:26:25>> I don't think so. If if anyone wants
1:26:27materials or anything, reach out. Uh but
1:26:29other than that
1:26:30>> absolutely
1:26:31>> it's been yeah pleasure. All the
1:26:32questions were really fantastic.
1:26:35>> Yeah. Thank you so much everyone for
1:26:37coming.
1:26:38>> Amazing. Thanks so much um everyone.
1:26:41We'll see you next week at the same time
1:26:44um for our for our third lecture in the
1:26:46series. So take care and have a great
1:26:47week.
1:26:50>> Bye everyone.

About this transcript

This page contains the full transcript of BREAD-IGC virtual PhD-level course 2026 | Lecture 2 History, Persistence, and Development by International Growth Centre, generated from the public captions YouTube serves with the video. The transcript has 14,592 words across 2,193 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.