2. Counting Words & Phrases - [E2] 2. Chen, Keith. 2013.

[JunsolKim]

Post questions here for this week's exemplary readings: Chen, Keith. 2013. “The Effect of Language on Economic Behavior: Evident from Savings Rates, Health Behaviors, and Retirement Assets.” American Economic Review 103(2): 690-731. (But you must also skim the retraction/qualification in Roberts, Seán G., James Winters, and Keith Chen. 2015. “Future Tense and Economic Decisions: Controlling for Cultural Evolution.” PLOS ONE 10(7): e0132145.)

[GabeNicholson]

Going into this paper, I was definitely quite skeptical of Chen's 2013 theory, for exactly the reason mentioned in the paper by Winters (2015): the prediction a-priori would be that cultures which condone saving would develop precise ways of grammatically differentiating the future and the present. Humans are known to discount the future, so how could it be that a more imprecise way of defining time would lead to better long-term planning? In the same way that we are prone to laziness, we usually have to go out of our way to make a strong habit of working out and staying healthy.

Which brings me to my general question, how can we defend against these potential spurious correlations in research when , by definition, there are an infinite amount of them that can arise? For cases such as these, should scholars preemptively require even lower p-values to account for the multiple testing that likely goes on but remains unstated in the methods section? It's an unfortunate conclusion, but I think the 33% FDR warrants some reconsiderations on how statistical testing is done in the social sciences.

[isaduan]

Going into this paper, I was definitely quite skeptical of Chen's 2013 theory, for exactly the reason mentioned in the paper by Winters (2015): the prediction a-priori would be that cultures which condone saving would develop precise ways of grammatically differentiating the future and the present. Humans are known to discount the future, so how could it be that a more imprecise way of defining time would lead to better long-term planning? In the same way that we are prone to laziness, we usually have to go out of our way to make a strong habit of working out and staying healthy.

Which brings me to my general question, how can we defend against these potential spurious correlations in research when , by definition, there are an infinite amount of them that can arise? For cases such as these, should scholars preemptively require even lower p-values to account for the multiple testing that likely goes on but remains unstated in the methods section? It's an unfortunate conclusion, but I think the 33% FDR warrants some reconsiderations on how statistical testing is done in the social sciences.

To do the author justice, the author does develop two mechanisms: (1) explicit future grammar makes people feel like future is more distant; (2) explicit future grammar leads to a more precise belief about the future rewards. These mechanisms seem psychological, instead of cultural & social. So one way to test the mechanism would be through experiments - within strong-FTR language, whether the presence of explicit future grammar structures leads people to be worse at future planning ?

[Hongkai040]

Going into this paper, I was definitely quite skeptical of Chen's 2013 theory, for exactly the reason mentioned in the paper by Winters (2015): the prediction a-priori would be that cultures which condone saving would develop precise ways of grammatically differentiating the future and the present. Humans are known to discount the future, so how could it be that a more imprecise way of defining time would lead to better long-term planning? In the same way that we are prone to laziness, we usually have to go out of our way to make a strong habit of working out and staying healthy.

Which brings me to my general question, how can we defend against these potential spurious correlations in research when , by definition, there are an infinite amount of them that can arise? For cases such as these, should scholars preemptively require even lower p-values to account for the multiple testing that likely goes on but remains unstated in the methods section? It's an unfortunate conclusion, but I think the 33% FDR warrants some reconsiderations on how statistical testing is done in the social sciences.

I didn't dig into the author's 2013 paper, but I find that the author did make concession in the abstract and the conclusion sections that "the possibility that language is not causing but rather reflecting deeper differences that drive savings behavior." FYI. For me, the most valuable part of this research is the idea is really informative. I'm rather interested in the criterions of choosing behaviors. From my perspective, many of them are modern behaviors and they are the natural results of education (and science!). Like, smoke is bad to your health, and using condoms is safe sex.Plus, Chen mentioned that FTR is very old in many languages. Does this mean that what Chen measures are values rooted in Strong-FDR and Weak-FDR cultures?

[weeyanghello]

It is very interesting for me whenever disciplines cross-pollinate. There are many strands and schools of thought within a discipline, especially in linguistics. Linguistics is far from being a homogenous discipline. And it seems that Chen has taken up precisely the strand of linguistics that is structuralist in its analysis of language. Chen cites many authors of the linguistic turn who make the distinction between langue (language as a system) and parole (language as speech), and they take up the former as a more fruitful site of study. Of the authors cited: Saussure and the school(s) of linguistic structuralism that he inspired, along with Chomsky, Steven Pinker, and the strong/weak interpretation of the Sapir-Whorf Hypothesis. (They also cite Roman Jakobson but did not go into the poetics of message that Jakobson is famous for in addition to the 6 functions of language, but I digress). Linguistic structuralism basically treats language as deconstruct-able into grammatical categories, and that these grammatical categories are responsible for influencing speech, human behavior, and even social reality. There is also less emphasis in linguistic structuralism on the pragmatics but more on the semantics of language as decontextualizable stretches of denotational code. Not going too far into the problems of linguistic structuralism, my question is what kinds of affinity computational analysis can have with traditional social sciences and humanities? And what debates or arguments do computational analyses such as Chen's implicitly and inadvertently wade into and support? Can we think about computational analysis without structualist implications (I am very invested in this one; after all this is the reason I am taking this course!)?

[chentian418]

This is a very inspiring paper about how to doing inference in social science subjects with the variables from linguistic measurements. However, as the author said, it's likely that language is not causing but rather reflecting deeper differences that drive savings behavior, and I believe it's challenging to rule out some culture values that are makers of the causal factors. I have two question about the measurements about future-time reference intensity:

1. I am curious about how to automate the process of identifying the number of verbs which are grammatically future-marked and the future-referring verbs.
2. Besides the dimension of "verb ratio" counts and "sentence ratio counts", I am think about are there any other measures about FTR intensity, for example, can we combine context and semantics before and after the verbs so that the measures can be contextualized for each language.

[YileC928]

The paper is indeed an innovative one and has triggered wide discussion ever since its publication. Adding on to @Halifaxi ’s question, I am also interested to know how we could potentially test the causal mechanisms between grammatical future-time references and intertemporal decision behaviors. Would experiment be a possible solution? I just read a recent paper (Chen, He, and Riyanto, 2019) which test the linguistic-savings hypothesis on Chinese speakers by manipulating sentences to be with/without future-time indication. While their result does not exhibit supporting evidence for the hypothesis, their setting may not be generalizable to other languages and longer time horizon. I am wondering what kind of sample and research design future projects may explore to further dig into this field.

[sborislo]

I find the evidence for the role of strong future-time reference languages in health and financial behavior compelling; however, Chen naturally did not control for some crucial factors (e.g., wealth inequality) which could help explain some of strong-FTR's effects. Therefore, I wonder why Chen did not provide another test of the role of FTR-type on these behaviors. For example, why did Chen not examine the history of languages in each region, which could offer a clearer picture of how language might shape culture? Could regions with more consistency in their FTR-type experience stronger effects on the measured behaviors?

(As a side question, I thought there were a few strange assumptions made in the analyses, including one Chen himself acknowledged yet still made (that those with higher incomes don't retire for longer, which I would say is obviously violated in this case). Why does Chen choose to make such strong assumptions?)

[XiaotongCui]

Very interesting research! However, I believe national savings are influenced by numerous factors, especially economic ones. As an Economics student, I don’t think income, GDP, and those factors alone can account for the variations caused by economic conditions...