Elsevier

Labour Economics

Volume 71, August 2021, 102028
Labour Economics

Heterogeneous effects of poverty on attention

https://doi.org/10.1016/j.labeco.2021.102028Get rights and content

Highlights

  • Analyzing cognitive test results before and after payday.

  • Application of the causal forest method to estimate heterogeneous treatment effects.

  • Poor financial circumstances impede attention of the low-income young and elderly.

  • Corroboration of the heterogeneous effects through an independent experiment.

Abstract

We examine heterogeneity in the effect of poor financial circumstances on attention. Our analysis uses data from an experiment, which randomly assigned low-income individuals to perform a cognitive test before or after payday. On average, and based on traditional subgroup analysis, the experiment did not suggest that the poorer financial circumstances before payday impeded cognitive function. Using the causal forest method, however, our heterogeneity analysis suggests that there are indeed detrimental effects among young and elderly individuals with very low incomes. We can confirm this finding in an independent experiment, using only traditional subgroup analysis.

Introduction

Many studies have documented associations between poverty and less beneficial behavior. For example, the poor are less likely than those with higher incomes to make use of preventive health services, and more likely to smoke cigarettes, play the lottery, and borrow more often at high cost.1 Despite long-standing debates in economics and other disciplines, the reasons for such behavior remain unclear and the topic itself controversial. One recent hypothesis has focused on the financial circumstances of the poor and the potentially detrimental impact of these on attentional focus: In a sample of farmers from India, Mani et al. (2013) found that participants showed reduced attentional performance before harvest, when poor, compared to after harvest, when rich. The authors suggested that a preoccupation with monetary concerns may leave the farmers before harvest with fewer mental resources available for other processes.2

In the only other study to have investigated this hypothesis empirically to date, Carvalho et al. (2016) assigned a sample of low-income US individuals randomly to perform a number of cognitive tests before or after payday. The individuals surveyed before payday faced poorer financial circumstances than those surveyed after payday. However, the authors found no before-after differences in cognitive function in the full sample or selected subgroups. These mixed empirical findings, and the dearth of studies on this hypothesis in general, highlight the need to identify, at a more detailed level, the groups of individuals in which poor financial circumstances might have detrimental effects on cognitive function.

To contribute to this area of study, we therefore analyze heterogeneity in the effect of financial circumstances on attention, focusing on identifying individuals in whom poorer financial circumstances have negative effects. To do so, we use data from the experiment conducted by Carvalho et al. (2016). For our heterogeneity analysis, we use the causal forest method by Athey et al. (2019), which was developed specifically to explore heterogeneous treatment effects in experiments. The method can be described as an adaptive nearest-neighbors approach that exploits ideas from the random forest machine learning literature to determine the relevant neighborhoods for estimating conditional average treatment effects at given points in the covariate space. Compared with traditional ordinary least squares (OLS) subgroup analyses, the causal forest method allows non-linear treatment effects to be estimated in a fully flexible way and circumvents the need to specify an interacted model, which may not always be straightforward (especially when the number of covariates is large). We examine effect heterogeneity using a rich set of 37 policy-relevant, pre-treatment covariates, including age, income, employment status, and measures of financial strain in the past. Our causal forest analysis proceeds in the following steps: First, we investigate which covariates are particularly relevant for heterogeneity in the treatment effect. Next, we examine how the effect varies across the most important variables. Subsequently, we study, in greater detail, the effect heterogeneity in regions of the covariate space where the previous step indicates particularly detrimental effects.

The results of our analysis suggest that there is strong effect heterogeneity in the two covariates age and income. For old and young individuals who received a very low income around the time of the experiment, we find that the poorer financial circumstances before payday had detrimental cognitive effects. We verify this finding using a second, independent, experiment conducted by Carvalho et al. (2016). Our results provide further evidence that there may be a causal effect of poverty on attention. They also demonstrate the benefit of using the causal forest method to identify treatment effect heterogeneity that may have been overlooked in traditional subgroup analyses.

The remainder of this paper is structured as follows. Section 2 describes the experiment and our analysis sample. Section 3 explains the causal forest method. Section 4 presents average effect estimates for the full sample, the results of our heterogeneity analysis, and investigates the findings of our heterogeneity analysis in an independent experiment. Section 5 concludes.

Section snippets

Experiment

Carvalho et al. (2016) conducted their experiment twice, once among members of the RAND American Life Panel and then again among members of the GfK KnowledgePanel. Both are ongoing online panels with individuals aged 18 and over living in the United States. The authors restricted the sample for each experiment to individuals with an annual household income of $40,000 or less. For our analysis, we use the data from the GfK KnowledgePanel because it had the larger sample size, and because its

Methodology

The goal of our analysis is to study heterogeneity in the effect on attention of poorer financial circumstances before payday. To do so, we estimate conditional average treatment effects using the causal forest method, which is based on the generalized random forest framework by Athey et al. (2019). The method is designed for studying treatment effect heterogeneity in experiments and can be described as an adaptive nearest-neighbors approach that uses a type of random forest technique to

Results

Section 4.1 describes the OLS average effect estimates for the full sample. Section 4.2 subsequently presents the results of our heterogeneity analysis, which uses subsample splitting to guard against spuriously extreme effects. Section 4.3 gives the estimates for our subgroup analysis based on the insights from the heterogeneity analysis, using our main analysis sample and an additional, independent, sample by Carvalho et al. (2016).

Conclusion

In this paper, we examine heterogeneity in the effect of financial circumstances on attention. Our analysis is based on data from an experiment by Carvalho et al. (2016), which randomly assigned low-income individuals in the US to perform a cognitive test before or after payday. To explore heterogeneity in the effect of the poorer financial circumstances before payday, we use the causal forest method by Athey et al. (2019), which is designed for studying heterogeneous treatment effects in

References (24)

  • J.S. Ross et al.

    Use of preventive care by the working poor in the United States

    Prev. Med.

    (2007)
  • J.M. Shapiro

    Is there a daily discount rate? Evidence from the food stamp nutrition cycle

    J. Public Econ.

    (2005)
  • S. Athey et al.

    Recursive partitioning for heterogeneous causal effects

    Proc. Natl. Acad. Sci.

    (2016)
  • S. Athey et al.

    Generalized random forests

    Ann. Stat.

    (2019)
  • M. Bertrand et al.

    A behavioral-economics view of poverty

    Am. Econ. Rev.

    (2004)
  • M. Bertrand et al.

    Behavioral economics and marketing in aid of decision making among the poor

    J. Public Policy Mark.

    (2006)
  • M.M. Botvinick et al.

    Conflict monitoring and cognitive control

    Psychol. Rev.

    (2001)
  • L. Breiman

    Random forests

    Mach. Learn.

    (2001)
  • L.S. Carvalho et al.

    Poverty and economic decision-making: evidence from changes in financial resources at payday

    Am. Econ. Rev.

    (2016)
  • Bourke, N., Horowitz, A., Roche, T., 2012. Payday Lending in America: Who Borrows, Where They Borrow, and Why. Pew...
  • Clotfelter, C. T., Cook, P. J., Edell, J. A., Moore, M., 1999. State Lotteries at the Turn of the Century: Report to...
  • S.R. Dube et al.

    Cigarette Smoking among Adults and Trends in Smoking Cessation – United States, 2008

    MMWR Morb. Mortal Wkly Rep. Ser.

    (2009)
  • Cited by (0)

    We thank Axel Börsch-Supan, Martin Huber, Mike Hurd, Jann Spiess, Stefan Wager, Frank Windmeijer, Joachim Winter, and participants of the Machine Learning in Economics and Econometrics Workshop in Munich 2018, the Annual Conference of the German Health Economics Association in Augsburg 2019, the 6th Annual Conference of the International Association for Applied Econometrics in Nicosia 2019, the 31st Conference of the Austro-Swiss Region of the International Biometric Society in Lausanne 2019, as well as seminars at Hamburg University, Harvard University, MEA and TU Braunschweig for helpful comments. We thank Tabea Braun for excellent research assistance. We acknowledge support from the German Research Foundation (DFG), Germany - 431701914. The authors declare that they have no relevant or material financial interests related to the research described in this paper.

    View full text