Defining Retirement in Administrative and Survey Data: An Example with Swedish Data

Haodong Qi, Stockholm University
Linda Kridahl, Department of Sociology, Stockholm University
Ann-Zofie Duvander, Demography Unit, Department of Sociology, Stockholm University

Retirement often refers to the process when older workers withdraw themselves from the labor market. The measure of retirement may not necessarily be consistent across different empirical analyses, which may restrict the comparability of the empirical results across different data sources, studies, countries, social context, and over time. The study’s aim is to scrutinize and compare different measures of retirement timing, and to create more clarity in how various measures affect the understanding of retirement patterns while focusing on gender, cohorts and period differences. We do this in two ways. First, we first explore how results on retirement patterns differ with alternative measures based on pension and labor income in Swedish administrative registers. Second, we compare administrative income-based measures with a self-assessed indicator from survey data. Preliminary results show that when retirement timing is defined as when the share of pension out of total income equals or exceeds 50%, the retirement timing spikes at age 65. However, this spike is shifted to age 66 with a definition of pension share equal or exceeding 80% of total income. Patterns are similar for women and men but the spike when using the 50% definition is sharper for women. The results imply that the age pattern of retirement is sensitive to different income-based measures. Using the smaller share of pension reveal an early retirement pattern, as more individuals will be classified as retirees at a younger age. Conversely, a large share result in a pattern of delayed retirement, because individuals at younger age will be “disqualified” as retirees. The results also show that the 50% threshold is more similar to self-assessed retirement age derived from survey data. The age profile is closely mirrored by the pattern using the 50% threshold. The study finds that empirical evidence that different retirement definitions are not comparable.

Understanding retirement behavior hasbecome increasingly crucial for an ageing society to sustain its transfersystem, including pension and taxation, and to provide adequate welfareservices, encompassing health care and elderly support. Retirement timing isalso highly associated with the health situation of the older population and istherefore important not just for projections of labor market participation, butalso of population projections. However, the timing of retirement is not alwaysa clear turning point in an individual’s life; it may as well be a gradualprocess. With the recent advances in collecting micro data, retirement studieshave been increasingly reliant on individual-level information. Whileretirement often refers to the process that older workers withdraw themselvesfrom the labor market, its measure is not necessarily consistent acrossdifferent empirical analyses. Inconsistent retirement measures may restrict thecomparability of the empirical results across different data sources, studies,countries, social context, and over time (Denton& Spencer, 2009).

The study’saim is to scrutinize and compare different measures of retirement timing, andto create more clarity in how various measures affect the understanding ofretirement patterns while focusing on differences across gender, cohorts andperiod. We do this in two ways. First, we explore how results on retirementpatterns differ with alternative measures based on pension and labor income in Swedishadministrative registers. Second, we compare administrative income-basedmeasures with a self-assessed indicator from survey data.

The study populationincludes cohorts born 1935-1945, observed between ages 55 to 67. In the firststep, we use generate different income-based measures of retirement and linkthem to various socio-economic and demographic characteristics using Swedishadministrative data. The measure is based on annual information on pension andlabor income. Pension income encompasses old-age public pension andearly retirement benefits. Labor income includes earned income as well asunemployment and sick leave compensation. We calculatethe share of pension income relative to the sum of pension and labor income,i.e. s=Pension/(Pension+labor income). Then we define retirement status basedon different values of s. When s=1, the individual is classified as fullyretired. In the first step we examine the thresholds 0.5 and 0.8, whichindicate that pension is 50 percent respectively 80 percent of the total incomea given year. In the second step, we compare the results from the register datawith self-assessed retirement status using the Swedish wave of the Generationand Gender Survey.

The followingresults are preliminary. Figure 1, shows the age pattern of retirement ratesfor both men and women with the share of pension relative to total income equalto or greater than 50% and 80%, respectively. It is clear that with twodifferent measures of retirement, the age-specific rates differ considerably.In particular, when retirement is classified by a low threshold, the ratespikes at age 65. However, when a high threshold is used, the spike ofretirement rate is shifted to age 66. This is true for both men and women butthe spike for women when the 50% measure is used, is sharper than the spike formen. Furthermore, Figure 2, shows the retirement rates calculated based on self-assessedretirement status in the Swedish Gender and Generation Survey. The age profileis closely mirrored by the pattern shown in Figure 1 using the threshold thatpension equals or exceeds 50% of total income. In particular, they bothindicate that the retirement rates spike at age 65 for both men and women.Moreover, preliminary discrete-time survival analyses withthe two measures of retirement (50% and 80%), indicate that the highest risk ofretirement is at ages 65 and 66, respectively, even after controlling forseveral demographic and socioeconomic characteristics.

Our tentativeconclusion based on the preliminary results is that the pattern of retirementis sensitive to what income-based measures are used to define retirementstatus. A small value of s tends to reveal an early retirement pattern, as moreindividuals will be classified as retirees at a younger age. Conversely, alarge value of s may result in a pattern of delayed retirement, simply becausemore individuals at younger age will be “disqualified” as retirees. We alsonoticed that the 50% threshold is more similar to self- assessed retirement agederived from survey data. In sum, the study finds empirical evidence that differentretirement definitions are not comparable.

References

Denton,F. T., & Spencer, B. G. (2009). What Is Retirement? A Review and Assessmentof Alternative Concepts and Measures. Canadian Journal on Aging / La RevueCanadienne Du Vieillissement, 28(1), 63–76.

 


Figure1.Age-Specific Retirement Rates, register data s > 0.5 and 0.8, cohorts1935-1945.

Figure 2. Age-SpecificRetirement Rates, survey data, cohorts 1935-1945, n: 1617.

 

Presented in Session 1234: Posters