The Age–Period–Cohort Problem in Hedonic House Prices Models

The age–period–cohort problem has been studied for decades but without resolution. There have been many suggested solutions to make the three effects estimable, but these solutions mostly exploit non-linear specifications. Yet, these approaches may suffer from misspecification or omitted variable bias. This paper is a practical-oriented study with an aim to empirically disentangle age–period–cohort effects by providing external information on the actual depreciation of housing structure rather than taking age as a proxy. It is based on appraisals of the improvement values of properties in New Zealand to estimate the age-depreciation effect. This research method provides a novel means of solving the identification problem of the age, period, and cohort trilemma. Based on about half a million housing transactions from 1990 to 2019 in the Auckland Region of New Zealand, the results show that traditional hedonic prices models using age and time dummy variables can result, ceteris paribus, in unreasonable positive depreciation rates. The use of the improvement values model can help improve the accuracy of home value assessment and reduce estimation biases. This method also has important practical implications for property valuations.

When less conditioning provides better estimates: overcontrol and endogenous selection biases in research on intergenerational mobility

The counterfactual approach to causality has become the dominant approach to understand causality in contemporary social science research. Whilst most sociologists are aware that unobserved, confounding variables may bias the estimates of causal effects (omitted variable bias), the threats of overcontrol and endogenous selection biases are less well known. In particular, widely used practices in research on intergenerational mobility are affected by these biases. I review four of these practices from the viewpoint of the counterfactual approach to causality and show why overcontrol and endogenous selection biases arise when these practices are implemented. I use data from the German Socio-Economic Panel Study (SOEP) to demonstrate the practical consequences of these biases for conclusions about intergenerational mobility. I conclude that future research on intergenerational mobility should reflect more upon the possibilities of bias introduced by conditioning on variables.

Child stature, maternal education, and early childhood development in Nigeria

Data from the 2016–17 Multiple Indicator Cluster Survey from Nigeria are used to study the relationship between child stature, mother’s years of education, and indicators of early childhood development (ECD). The relationships are contrasted between two empirical approaches: the conventional approach whereby control variables are selected in an ad-hoc manner, and the double machine-learning (DML) approach that employs data-driven methods to select controls from a much wider set of variables and thus reducing potential omitted variable bias. Overall, the analysis confirms that maternal education and the incidence of chronic malnutrition have a significant direct effect on measures of early childhood development. The point estimates based on the ad-hoc specification tend to be larger in absolute value than those based on the DML specification. Frequently, the point estimates based on the ad-hoc specification fall inside the confidence interval of the DML point estimates, suggesting that in these cases the omitted variable bias is not serious enough to prevent making causal inferences based on the ad-hoc specification. However, there are instances where the omitted variable bias is sufficiently large for the ad hoc specification to yield a statistically significant relationship when in fact the more robust DML specification suggests there is none. The DML approach also reveals a more complex picture that highlights the role of context. In rural areas, mother’s education affects early childhood development both directly and indirectly through its impact on the nutritional status of both older and younger children. In contrast, in urban areas, where the average level of maternal education is much higher, increases in a mother’s education have only a direct effect on child ECD measures but no indirect effect through child nutrition. Thus, DML provides a practical and feasible approach to reducing threats to internal validity for robust inferences and policy design based on observational data.

Can Panel Data Methodologies Determine the Impact of Climate Change on Economic Growth?

Several major papers have been published over the last ten years claiming to have detected the impact of either annual variations in weather or climate change on the GDPs of most countries in the world using panel data-based statistical methodologies. These papers rely on various multivariate regression equations which include the annual average temperatures for most countries in the world as one or more of the independent variables, where the usual dependent variable is the change in annual GDP for each country from one year to the next year over 30-50 year time periods. Unfortunately, the quantitative estimates derived in these papers are misleading because the equations from which they are calculated are wrong. The major reason the resulting regression equations are wrong is because they do not include any of the appropriate and usual economic factors or variables which are likely to be able to explain changes in GDP/economic growth whether or not climate change has already impacted each country’s economy. These equations, in short, exhibit suffer from “omitted variable bias,” to use statistical terminology.

Love Thy Neighbour? Perceived Community Abidance and Private Compliance to COVID-19 Norms in India

To arrest the spread of COVID-19 infection, strict adherence to frequent hand washing and respiratory hygiene protocols have been recommended. While these measures involve private effort, they provide health gains along with collective community benefits and hence are likely to be driven by pro-social motives like altruism and reciprocity. Using data from 934 respondents collected from April till May 2020 across India, we assess if changes in perceived community compliance can predict changes in individual compliance behaviour. We observe statistically significant and positive relationship between the two, even after accounting for observable and omitted variable bias allowing us to view the results from a plausible causal lens. Further, we find subsequent lockdowns having a detrimental effect on individual compliance though the gains from higher perceived community compliance seem to offset this loss. We also find positive perceptions about community can be particularly effective for people with pre-existing co-morbidities. Our findings underscore the need for multi-level behavioural interventions involving local actors and community institutions to sustain private compliance during the pandemic. We suggest these interventions need to be specially targeted for individuals with chronic ailments and emphasize on community behavioural practices in public messaging. JEL Codes: I12, I18, I19, I31

Doctoral Education, Job Mismatch, and Wage Consequences in South Korea: A Propensity Score Matching Approach

This study extends the previous literature on the wage effects of over-education, focusing on young doctorate holders (DHs). It also contributes to the conventional over-education literature on a causal relationship between over-education and wages by implementing techniques of propensity score matching (PSM). By tackling potential bias as a consequence of omitted variable bias via the PSM strategy, this study provides evidence of the negative influence of over-education on wages (i.e., the over-education wage penalty) once potential sources of bias are adequately considered. While the current analysis is focused on one country, South Korea, its results might be relevant for many other countries that have experienced a rapid expansion in the supply of DHs over recent years.

Omitted Variable Bias of Lasso-Based Inference Methods: A Finite Sample Analysis

We study the finite sample behavior of Lasso-based inference methods such as post double Lasso and debiased Lasso. We show that these methods can exhibit substantial omitted variable biases (OVBs) due to Lasso not selecting relevant controls. This phenomenon can occur even when the coeffcients are sparse and the sample size is large and larger than the number of controls. Therefore, relying on the existing asymptotic inference theory can be problematic in empirical applications. We compare the Lasso-based inference methods to modern highdimensional OLS-based methods and provide practical guidance.

The Political Premium of Television Celebrity

This paper studies the electoral consequences of television stardom through the career of Ronald Reagan. I utilize quasi-experimental variation in television reception to estimate the causal effect of celebrity exposure on political support. I find that Reagan’s tenure as the host of a 1950s entertainment television program translated into support for his candidacy, in terms of votes and political donations, nearly two decades after the show’s first airing. Placebo checks suggest that this impact is not driven by unobserved heterogeneity or omitted variable bias. The effect was especially pronounced in the 1976 Republican primary elections relative to the general presidential elections and partially dissipated in locations where Reagan was a known political entity. Using the American National Election Studies (ANES) surveys, I provide evidence on possible mechanisms. Consistent with rational updating, nonpolitical media increased voters’ assessment of Reagan’s character and leadership, personalizing political considerations in elections featuring him. (JEL D72, L82, Z13)