CapMod: A Simulated Society to Evaluate Empirical Estimators of Capabilities

ABSTRACT

This article introduces an innovative approach to the validation of empirical methods aiming at estimating capabilities. Validating these empirical methods is difficult because capabilities are not directly observable. We propose a computational model to generate data from a simulated society, where we observe both functionings and capabilities. These data can then be used to compare estimation methods against each other. Most importantly, our data generating process is completely disconnected from any estimation procedure and can, therefore, be used to compare a variety of methods. The model is calibrated to the Mexican economy and coherently reproduces many stylised facts of this economy. The article also proposes a short illustrative analysis on how to use the simulated data and a section on how this can be implemented practically.

KEYWORDS:

• Capabilities
• Simulated society
• Agent-based model
• Measurement
• Methodology
• Model validation

JEL CLASSIFICATIONS:

• C15
• C52
• C63
• D63
• I00
• I20

Acknowledgments

We would like to thank participants at the 2018 HDCA conference in Buenos Aires for helpful comments.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 Gilbert (2008) and Gilbert and Troitzsch (2005) provide excellent introductions to agent-based modelling. For a short introduction to agent-based models in the context of economic development, see Chávez-Juárez (2017).

2 See Krishnakumar (2017) for a review of proposed methods.

3 We distinguish a total of 8 levels, but for the sake of readability, we only highlight four levels in these graphs.

4 In Section 4.1, we discuss the performance of the priority-approach by comparing the consumption pattern to actual data from a Mexican household survey. Chávez-Juárez, Trujillo, and Ortega (2018) use a very similar approach and analyse the results in much more details. They find that the priority approach produces results that respect a wide range of stylised facts found in consumption data. Gómez (2019) performed a simulation exercise comparing four different consumption theories and found that the priority approach was best performing along with another theory, both outperforming a standard utility maximisation approach.

5 We acknowledge that in the real world the social network also develops in the neighbourhood or at the workplace. We do not include co-workers because we generally focus on young individuals for our exercise (see Section 4 and we ignore neighbourhood effects to keep the model simple and manageable.

6 As mentioned earlier, through changes to the parameters CapMod is able to generate different types of societies. This is important because not all empirical models to estimate capabilities might work for all types of societies. For instance, methods based on simple averages of functionings could work well in a setting like ours where capabilities and functioning are quite related but fail to estimate capabilities well for societies where people cannot reach their capabilities. Such exercises can easily be implemented using CapMod with different settings. For the sake of space, we present here only the results of one setting.

7 This is not a Monte Carlo simulation in the econometric sense as described in the introduction. We run CapMod 25 times and obtain 25 samples from a data-generating process with no direct connection to any of the analysed operationalisation approaches.

8 For a more detailed discussion of the structural equation model (SEM) approach, please refer to Chávez-Juárez and Krishnakumar (2018).

9 We are currently setting up a website of the project where interested researchers can find all relevant information (source code, related publications, data, tutorials, etc.).

10 See https://github.com/fwchj/capmod/ for more details.

11 See https://github.com/fwchj/capmod/ for more details.

Additional information

Notes on contributors

Florian Chávez-Juárez

Florian Chávez-Juárez is a co-founder and managing director of CORESO—Collaborative Research Solutions in Switzerland and associate researcher at the National Laboratory of Public Policy in Mexico City. Previously he was assistant at the National Laboratory of Public Policy (LNPP) at the Centre for Center for Research and Teaching in Economics (CIDE) in Mexico City and head of LNPP's Simulation Unit. His main research interests include health economics, inequality of opportunity, capability approach, agent-based modelling and public policy simulation.

Jaya Krishnakumar

Jaya Krishnakumar is a full professor of Econometrics at the University of Geneva, Switzerland. She is also a visiting professor at the Indian Institute of Management, Bangalore, and Madras School of Economics, Chennai, India. Her research interests include panel data econometrics, multivariate models with latent variables and quantitative methods for multi-dimensional well-being analysis. She has publications in leading international econometrics/economics journals for example in Econometric Theory, Journal of Econometrics, Oxford Bulletin of Economics and Statistics, Journal of Public Economics, European Economic Review, Health Economics, and World Development. She has also edited and contributed chapters in books in Econometrics and on the Capability Approach. She is a member of the Advisory Panel for the Human Development Reports of the UNDP, and a Fellow of the Human Development and Capabilities Association. She has also been a member of the academic experts panel for World Bank's Women, Business and The Law Index 2019, as well as an Advisor for the SDG Action Manager launched by B-Lab along with the UN Global Compact in early 2020.