Measuring Social Mobility in Longitudinal Data-Scarce Settings

Social mobility covers a wide scope and can be viewed from different perspectives. The Key Indicators 2022 report notes that “intragenerational mobility measures how a person’s socioeconomic status changes over his or her lifetime, while intergenerational mobility compares an adult’s status with that of his or her parents. Social mobility may also be characterized as either ‘relative’ or ‘absolute’ movements between socioeconomic status levels. Relative social mobility refers to a person’s ranking with respect to a specific socioeconomic hierarchy (such as income deciles), so that a person ranking higher over time means that another person has ranked lower. Absolute social mobility ignores rankings and simply assesses whether a person has changed socioeconomic status over time, compared to a predefined level.” 

Regardless of perspectives or the lens being used, a number of metrics of social mobility requires longitudinal data that tracks how people move from one socioeconomic status to another over time. However, a number of developing economies do not routinely collect suitable panel data and this is one of the factors contributing to the dearth of social mobility studies in a number of low- to middle-income country settings.

If the objective is to measure intergenerational social mobility, cross-sectional surveys can be used to collect retrospective socioeconomic information about the adult respondent’s parent that is easy to recall, say, educational attainment. In such case, social mobility could be measured based on correlation or difference between an adult respondent’s educational attainment with that of his or her parents. The data compiled in the World Bank’s Intergenerational Social Mobility Database follow this approach. 

Figure 1 shows that education mobility among people from lower socioeconomic classes was low in most economies in Asia and the Pacific, based on estimates using data drawn from the World Bank. In fact, only a few economies saw at least 20% of their 1980s cohort coming from the bottom half in education climb up to the top quartile.

Figure 1: Education Mobility of Those Born into the Bottom Half in Education, by Economy

Lao PDR = Lao People’s Democratic Republic, PRC = People’s Republic of China.
Notes: The figure represents 38 Asian Development Bank (ADB) member economies with available data. ADB placed on hold its assistance in Afghanistan effective 15 August 2021 and in Myanmar effective 1 February 2021. ADB did not make any consultations with either Afghanistan or Myanmar for the data in Figure 1. Educational attainment refers to the highest educational level completed among the following five categories based on the International Standard Classification of Education (ISCED): less than primary (ISCED 0), primary (ISCED 1), lower secondary (ISCED 2), upper secondary or postsecondary nontertiary (ISCED 3–4), and tertiary (ISCED 5–8). 
Source: Asian Development Bank estimates using data from World Bank’s Global Database of Intergenerational Mobility (accessed 1 February 2022).

However, collecting retrospective information about parents’ income or occupation, or even one’s own income in the past may be arduous to recall accurately. Hence, repeated cross sectional survey data facilitates compilation of aggregate measures of income or occupational distribution. Since not the same set of respondents is tracked over time, cross-sectional surveys cannot readily facilitate calculation of indicators, such as proportion of initially poor (nonpoor) people who became nonpoor (poor) later on, or proportion of initially poor (nonpoor) people who remained poor (nonpoor), or other measures of social mobility based on income or occupation.

Given the lack of suitable panel data in developing economies, there is growing interest in developing techniques for constructing synthetic or pseudo-panel data from repeated cross-sectional surveys, which can provide a more dynamic perspective of the evolution of socioeconomic status.

For illustration purposes, suppose there are two time periods of interest, time t and t+n, and in each survey round, there are four respondents. In a panel or longitudinal survey, respondents interviewed at time t, say A, B, C, and D, are re-interviewed at time t+n. In repeated cross-sectional survey data, a different set of respondents may be interviewed at time t+n as long as they have similar characteristics as A, B, C, and D.

If the main objective is to measure a metric of social mobility, such as  intergenerational income elasticity, pseudo-panel techniques, which involve grouping respondents of each survey into birth cohort groups, may be adopted (see Deaton, 1985; Antman and McKenzie, 2007). Aggregate data from formed cohort groups can then be the basis of calculating intergenerational income elasticity. 

However, such an approach of transforming individual or household-level data and expressing them as cohort averages may not work effectively if the objective is to measure movements into or out of poverty over time. In response, other researchers (Dang et al., 2014) proposed the creation of pseudo-panels by predicting what the income or consumption of persons A, B, C, and D could have been if they were interviewed at time t+n using a regression framework based on time-invariant characteristics.

Figure 2: Difference between Panel and Repeated Cross-Sectional Survey Data

Source: Authors’ visualization.

Using an actual panel data set, the performance of different pseudo-panel techniques in measuring a wide range of social mobility metrics was evaluated by Martinez et al. (2013). Results suggest that methods with more flexible income model specifications perform better than those with highly parameterized models. Garcés-Urzainqui et al. (2021) also provides a detailed review of alternative pseudo-panel methods, particularly in the context of studying poverty dynamics.

One of the challenges in providing an accurate measure of the long-term impact of the pandemic on poverty is the lack of suitable detailed data over time. Even a cross-sectional survey of household income and expenditure is typically conducted every 3 to 5 years and there is also a considerable lag before data become available. Not surprisingly, there is greater scarcity of longitudinal or panel data, which are more costly and complex to collect than cross-sectional surveys. 

Nonetheless, there is a need to probe the impact of the pandemic beyond examination of aggregate measures of poverty and inequality derived from multiple rounds of cross-sectional surveys to understand how long-term social mobility prospects of the poor and disadvantaged have been affected. In this regard, pseudo-panel methods may prove useful.

Additional details on social mobility in Developing Asia are available in the 2022 edition of the Key Indicators for Asia and the Pacific.

Asian Development Bank. 2022. Key Indicators for Asia and the Pacific 2022. Manila.

A. Martinez Jr. et al. 2013. Measuring Income Mobility using Pseudo-Panel Data. The Philippine Statistician. 62 (2). pp.71–99. 

D. Garcés-Urzainqui, P. Lanjouw, and G. Rongen. 2021. Constructing Synthetic Panels for the Purpose of Studying Poverty Dynamics: A Primer.  Review of Development Economics. 25 (4). pp.1803–1815. 

F. Antman and D.J. McKenzie. 2007. Earnings Mobility and Measurement Error: A Pseudo‐Panel Approach.Economic Development and Cultural Change. 56 (1).  pp. 125–161. 

H. Dang et al. 2014. Using Repeated Cross-Sections to Explore Movements Into and Out of Poverty. Journal of Development Economics. Vol. 107. pp. 112–128.