The Development Gap in Economic Rationality of Future Elites

Authors:

• Alexander W. Cappelen
• Shachar Kariv
• Erik Ø. Sørensen (contactperson for code and data, erik.sorensen@nhh.no)
• Bertil Tungodden

Abstract: We test the touchstones of economic rationality---utility maximization, stochastic dominance, and expected-utility maximization---of elite students in the U.S. and in Africa. The choices of most students in both samples are generally rationalizable, but the U.S. students' scores are substantially higher. Nevertheless, the development gap in economic rationality between these future elites is much smaller than the difference in performance on a canonical cognitive ability test, often used as a proxy for economic decision-making ability in studies of economic development and growth. We argue for the importance of including consistency with economic rationality in studies of decision-making ability.

Overview

The master file this replication package will:

1. Install the required versions of the necessary R packages from CRAN.
2. Downloads the necessary datafiles from Harvard Dataverse: https://doi.org/10.7910/DVN/CCODET .
3. Create all the displays in the paper as separate files (documented below).
4. Create markdown documents for numbers referenced in the paper but not explicitly part of produced tables.

Data availability statement

The experimental data used to support the findings of this study were collected by the authors. All data, with documentation, have been deposited in the public domain at Harvard Dataverse:

The experimental and survey data used to support the findings of this study have been deposited in the Harvard Dataverse repository (https://doi.org/10.7910/DVN/CCODET). The data were collected by the authors, are available under a Creative Commons Non-commercial license, and were assembled and prepared by code available at .

We certify that the author(s) of the manuscript have legitimate access to and permission to use the data used in this manuscript, and the data are licensed under a Creative Commons/CC0 license. See LICENSE_CC0.txt for details.

The data file is downloaded when the targets plan is first run.

Computational requirements

Software requirements

The analysis code was developed on an Ubuntu 22.04.2 machine with R 4.3.1. The script that runs all the code (main.R) will install the packages needed, of the correct version, into a local library using the renv library (which must be installed ahead of time). The list of packages (and all necessary recursive dependencies) is found in the renv configuration file renv.lock which should not need to be touched.

Memory and runtime requirements

Calculating the revealed preference statistics (in particular the CCEI for Expected Utility) is slow. the current setup parallelizes and runs separate branches for each participant. Running 32 processes in parallel on a modern server (AMD EPYC 7543P 32-Core Processor, 2.8GHz), the total time for calculating everything from scratch is about 2 hours (using less than 100GB memory). The parallel computations are controlled by the future library, and in main.R there is a line to control how many worker processes to start:

tarmakefuture(workers = 32)

Set the number of workers to a number that is compatible with the number of cores you can set aside.

Given the revealed preference statistics (from a previous run), the time needed to reproduce the analyses is trivial (less than a minute), the targets library caches data and precalculated results.

Description of programs/code

The graphical displays are produced in the graphs/ directory (as pdf-files). The tables are produced in the tables/ directory (as tex-files).

The file R/PollisonEtAl.R contains functions extracted from the replication package of Pollison et al (2020a,2020b). This code was published CC-BY 4.0. The code in R/functions.R contains an interface function calculate_rp_statistics(d) that calculates the revealed preference statistics we use on the subset of data d.

The dependencies of the analysis is controlled by the targets library. The list in _targets.R defines a directed acyclic graph of dependencies, and the tar_make command figures out which results are cached and which needs to be recalculated.

Running all the analysis should be possible from the command line with:

Rscript main.R

This will generate estimates.html at the root, and the displays in graphs/ (figures as pdfs) and tables/ (tables as tex-files).

The targets system is smart about caching intermediate results, so while running main.R takes a considerable amount of time for the first run, minor adjustments to the output routines in the vignettes are do not require the heavy computations to be re-run, and running main.R for the second time is almost free of costs with respect to changes in the display layer.

License for Code

The code in targets.R, R/functions.R and estimates.Rmd is licensed under a BSD-3-Clause license. See LICENSE_BSD-3-Clause.txt for details.

The code in R/PollisonEtAl.R is a collection of code from Pollison et al (2020b), covered by the Creative Commons BY 4.0 license. Apart from collecting several functions into one file, no change was made to this code.

Display items

The hard computational outcomes (the Revealed Preference statistics, in particular $e^{**}$) are precalculated as targets defined in the _targets.R file. The presentation layer is found in the estimates.Rmd file, which loads the targets pre-calculated and creates the display items and the statistics referenced in the text. The location is indicated by the name of the R-markdown *chunk within estimates.Rmd.

For the paper:

| Display Item | Filename | Chunk-name | Comment | |------------------|----------------------|------------------|--------------------------| | Figure 1 | na | na | theory, no data | | Figure 2 | graphs/Figure2.pdf | meandifferences | | | Figure 3 | graphs/Figure3.pdf | survivalgraph | | | Figure 4 | graphs/Figure4.pdf | zmeandifferences| | Table 1 | tables/developmentgap.tex | development_gap | minimal manual formatting added |

For the online appendix:

| Display Item | Filename | Chunk-name | Comment | |------------------|----------------------|------------------|--------------------------| | Figure A1 | graphs/cdfoutcomes.pdf | cdfoutcomes | | | Figure A2 | graphs/governmentpreference.pdf | governmentpreference | | | Figure A3 | graphs/happinessdistribution.pdf | happinessdistribution | | Table A1 | tables/studysubjects.tex | studysubjects | | Table A2 | tables/budgetshares.tex | budgetshares | | Table A3 | tables/developmentgapwithrisk.tex | developmentgapwithrisk | | Table A4 | tables/treatmenteffectriskaversion.tex | treatmenteffectriskaversion| | Table A5 | tables/treatmenteffectstakes.tex | treatmenteffectstakes | | Table A6 | tables/developmentgap.tex | developmentgap | Selection of rows |

References

• Polisson, Matthew, John K.-H. Quah, and Ludovic Renou (2020a). "Revealed Preferences over Risk and Uncertainty." American Economic Review 110(6): 1782-1820. https://doi.org/10.1257/aer.20180210.
• Polisson, Matthew, John K.-H. Quah, and Ludovic Renou (2020b). Data and Code for: Revealed Preferences over Risk and Uncertainty. Nashville, TN: American Economic Association [publisher], Ann Arbor, MI: Inter-university Consortium for Political and Social Research [distributor], 2020-05-27. https://doi.org/10.3886/E112146V1