Panel Surveys Research Paper

A widely used alternative term for panel survey is longitudinal survey. However, it is useful to use ‘longitudinal’ only in the context of longitudinal data; such data may be collected by a panel survey, but they could also come from retrospective reports in a cross-sectional survey.

Most panel surveys follow their sample elements for the duration of the survey. Other panel designs include a rotating panel survey, in which sample elements have a restricted panel life; as they leave the panel, new elements are added. Examples of this are the monthly US Current Population Survey (US Bureau of Labor Statistics 2000a) and the Canadian Labour Force Survey (Statistics Canada 2000), in which the sample at each month is composed of a number of rotation groups, each of which is a self-contained sample from the population of interest. One rotation group drops out and one rotation group enters each month. A split panel survey is a combination of a panel and a repeated survey (Kish 1986).

Panel surveys can differ markedly in the interval between rounds of data collection and the overall length of the survey. On the one hand, a consumer panel in market research may involve weekly contacts with respondents to collect their diary records of purchases. On the other hand, there are long-term panel surveys that may collect data only annually or even less frequently, and these surveys may continue for many years. Examples are the Panel Study of Income Dynamics (PSID), which had been collecting data on US households annually (biannually since 1997) since 1968 (Hill 1992), and the British National Child Development Study, which started in 1958 and has collected data on the sample persons on five occasions in the 33 years to 1991 (Institute of Education 2000).

Some panel surveys are concerned with population subgroups that have experienced the same event during the same time period, such as attending eighth grade in a given year, having been born in a particular week, or having been married during the same period. Since these sample universes are defined as cohorts, such studies are often called cohort studies. Examples include the various cohorts defined by age and sex in the US National Longitudinal Surveys (US Bureau of Labor Statistics 2000b). A limitation of cohort studies is that their results relate formally only to the particular cohorts from which the sample was selected. An extension of the approach is the multiple cohort design, in which several cohorts are followed in similar panel surveys. Thus, the National Longitudinal Surveys have begun three youth cohort panel surveys, in 1968 (males) and 1968 (females), 1979, and 1997, while the Monitoring the Future studies have selected samples of US high school seniors each year, and follow each sample in a panel survey (Monitoring the Future 2000).

Despite their lack of formal population representation, still other panel surveys are valuable because of their detailed interviews and often very long durations. Examples include Elder’s (1999) 30-year follow-up of individuals who had attended Oakland, California elementary schools and were enrolled in the Oakland Growth Study when in fifth and sixth grades and Furstenberg’s (Furstenberg et al. 1987) 30-year panel study of Baltimore teen mothers who had enrolled in a teen parenting program in 1965.

1. Uses Of Panel Surveys

To appreciate the analytic advantages of panel surveys, it is important to realize that the time dimension complicates analyses because: (a) characteristics of population elements may change over time, for instance, an individual may be single one month and married the next; and (b) the population itself may change in composition, with both sample units entering into and exiting from the population.

These complications, especially the first, give rise to a variety of objectives for analysis (Kish 1986, Duncan and Kalton 1987), including:

(a) Estimates of population parameters (e.g., monthly rates of unemployment, public opinion regarding gun control) at distinct time points. The resulting time series of such estimates may be the input for further analysis; see, for example, Smith (1994).

(b) Measurement of net change, defined as change at the aggregate level. An example is the change in the UK unemployment rate between two consecutive months or change in US public opinion regarding gun control before and after a gun-related event.

(c) Measurement of various components of individual change, including: (i) gross change—change at the individual level between two time points (e.g., relating the poverty status of sample members in a given year to their poverty status ten years later); (ii) average change, or trend, for each individual over a period of time, with the average serving to smooth out the effects of isolated perturbations and measurement errors; or (iii) instability for each individual (e.g., the stability of locus of control over 25 consecutive weeks from a sample of elderly in Eizenman et al. 1997).

(d) Aggregation of data for individuals over time. An example is the collection of incomes on an annual basis to relate childhood-long economic well-being to children’s developmental outcomes (Blau 1999).

(e) Measuring the frequency, timing, and duration of events occurring within a given time period. Examples are the proportion of persons who were ill in the past two weeks and the average duration of their illnesses, and the complete fertility histories of postmenopausal women.

If the composition of a population is sufficiently stable to be treated as static for the time period of interest, it may be possible to collect the data required to meet all of these objectives from a single cross-sectional survey conducted at the end of the period. Crucial here is that errors of measurement are acceptably small. Unfortunately, respondents’ memories are notoriously inadequate when a substantial time is involved or when the characteristics or events have a low degree of salience for the respondents (Bound et al. 2000). For example, respondents are often unable to recall important events such as unemployment (Mathiowetz and Duncan 1984) or hospitalizations (Cannell et al. 1981) within periods as short as several months, and they cannot reliably recall subjective experiences (Moss and Goldstein 1979). Thus, event data often need to be collected at several points throughout the period.

Repeated surveys can meet some of these analytic goals. Their particular strength is that at each round of data collection they routinely select a sample of the population existing at that time. They thus provide a series of cross-sectional estimates that can be used for objective (a) above. They can also be used to measure overall net change (b). The major limitation of a repeated survey is that it does not yield data to satisfy objectives (c) and (d): since elements are not explicitly included in several rounds, none of the components of individual change can be measured and individual data cannot be aggregated across rounds. Data on the frequency and timing of events in a specified period, as in objective (e), can be collected in a repeated survey, but with the measurement error problems noted above.

The major advantage of a panel survey over a repeated survey is its much greater analytical potential. It enables components of individual change to be measured, objective (c), and also the summation of a variable across time, objective (d). By providing high-quality measurements of the frequency, timing, and duration of events, panel surveys provide data useful for the estimating of event-history models (Tuma and Hannan 1984, Allison 1982, Yamaguchi 1991).

A panel survey can be much more efficient than a repeated survey for measuring net change, objective (b). Let the net change be y₂ – y₁, where y₁ and y₂ are the means of the variable of interest at times 1 and 2. Then the variance of the net change is V(y₂ – y₁) = V(y₁) + V(y₂) – 2ρ[V(y₁) V(y₂)]¹/² where ρ is the correlation between y₁ and y₂. In a repeated survey with two independent samples, ρ = 0, but in a panel survey ρ is the correlation between an individual’s y-values on the two occasions, which is often quite high. In this case, a panel survey will yield a much more precise estimate of net change, objective (b), than will a repeated survey of the same size. On the other hand, as discussed below, panel nonresponse between times 1 and 2 may result in more bias in the estimation of y in a panel than repeated survey.

A panel survey also permits the collection of a wider range of data on each sampled element than is possible with a repeated survey. The amount of data collected from any one respondent on one occasion is limited by the burden imposed; however, different data can be collected on different rounds of a panel survey, thus providing the possibility of accumulating an extremely wide range of variables for each sampled element. In the US Survey of Income and Program Participation (SIPP), for instance, core data on income and program participation are collected on each round, and additional data on different topics (e.g., health care utilization and financing, pension and retirement issues, housing conditions) are collected on specific rounds only (US Bureau of the Census 2000).

Some long-term panels have important intergenerational components, with data on children and their family contexts taken in early waves and measurements of the adult attainments of the children taken in later waves. Examples include population panels such as the National Longitudinal Survey youth panels, the PSID, the British Household Panel Survey (Institute for Economic and Social Research 2000), the German Socio-Economic Panel, and more specialized data collections such as the Oakland Growth Study (Elder 1999).

2. Problems With Panel Surveys

By attempting repeated interviews with the same sample, panel surveys have problems not found in single or repeated cross-sectional designs, the most important of which is panel nonresponse (initial-wave respondents may not respond in later waves). An additional potential problem with panel surveys is panel conditioning, where responses in a given interviewing round may be conditioned by participation in prior rounds of interviews.

Methods developed to cope with nonresponse bias include minimizing nonresponse in panel surveys and developing statistical adjustments for existing nonresponse. Existing panel surveys typically devote sizeable resources to maintaining high response rates, and sometimes are quite successful. For example, the National Longitudinal Survey of Youth conducted interviews in 1991 with 89 percent of the respondents in its initial, 1979, interview (MaCurdy et al. 1998). Losses in the British National Survey of Health and Development amounted to only 12 percent after 26 years (Atkins et al. 1981).

Incentive payments, respondent reports, persuasion letters, using administrative data for tracing, and collecting extensive contact information (e.g., on friends and relatives not living in the household who would know of address and telephone number changes) help minimize these problems (Freedman et al. 1980, Clarridge et al. 1978, Call et al. 1982).

As with any survey, sizeable nonresponse in a panel survey gives rise to concerns about nonresponse bias. The situation with the first wave of a panel survey corresponds to that with a cross-sectional survey in that very limited information is available on the nonrespondents. The situation with later wave nonresponse in a panel survey is, however, different: in this case a good deal of information is available about later wave nonrespondents from their responses on earlier waves. The earlier wave information can be used to investigate the possibility of nonresponse bias and to develop imputation and weighting nonresponse adjustments that attempt to reduce the bias (Kalton 1986, Lepkowski 1989).

With regard to conditioning, there is ample evidence from several surveys that initial responses in a panel survey differ substantially from those given in sub- sequent waves (Bailar 1975, 1979, Ghangurde 1982). In the case of the US Current Population Survey, estimates of unemployment from households entering the sample for the first time are almost 10 percent larger than the average over all eight monthly reporting periods. It is not clear whether there is more response bias in the initial or subsequent waves, because the repeated contact with respondents has ambiguous effects on the quality of the data. The crucial question, as yet unanswered for most phenomena reported in surveys, is whether it is merely the reporting of behavior or the behavior itself that is affected by panel membership.

It may be that data collected in subsequent panel waves is less biased, because repeated contact increases the probability that respondents understand the purposes of the study and are thus increasingly motivated to make the effort necessary to give more accurate answers. On the other hand, there is evidence from a validation study (Traugott and Katosh 1979) that extended participation in a panel study on election behavior not only increased the accuracy of responses on voting behavior but may indeed have increased the amount of voting, so that the behavior of the panel was no longer representative of the behavior of the population at large.

It seems unlikely that panel participation has pervasive behavioral effects, especially when changes in the behavior under investigation require more effort than making a trip to the polls. For example, economic behavior such as work effort, saving, commuting, and home ownership are all unlikely to be affected by responses to occasional interviews. Responses to attitudinal questions may be affected by panel membership if participation stimulates interest in the subject matter of the survey.

The limited membership in a rotating panel acts to reduce the problems of panel conditioning and panel loss in comparison with a nonrotating panel survey, and the continual introduction of new samples helps to maintain an up-to-date sample of a changing population. Rotating panels are used primarily for the estimation of cross-sectional parameters, objective (a), for the estimation of average values of population parameters across a period of time, objective (b), and for measuring net changes, objective (c). A rotating panel survey will generally provide more precise estimates of point of time and, especially, of change parameters than a repeated survey of the same size. Moreover, a rotating panel survey will sometimes have a cost advantage over a repeated survey. This will occur when it is cheaper to conduct a reinterview than an initial interview, as for instance is the case in the US Current Population Survey where initial interviews must be conducted by personal visit whereas reinterviews on some waves may be conducted by telephone (US Bureau of the Census 1978).

The ability of a rotating panel survey to measure components of individual change, objective (c), and to aggregate data for individuals across time, objective (d), is clearly restricted. Since rotating panels are not intended to serve these objectives, they can be designed to avoid the heavy expense of following movers that occurs with nonrotating panel surveys. Thus, for instance, the Current Population Survey employs dwellings, not households or persons, as the sampled units, so that there is no need to follow households or persons moving between panel waves.

In a split panel survey, the panel survey component can be used to measure components of individual change, objective (c), and to aggregate data for individuals over time, objective (d). Its permanent overlap aids in the estimation of net change, objective (b), between any two waves whereas the overlap in a rotating panel survey aids only in the estimation of net change between certain prespecified waves.

Both rotating and split panel survey designs provide samples of new entrants to the population and the capacity to use their panel survey components to check on biases from panel conditioning and respondent losses.

3. The Problem Of Changing Populations

The composition of almost every population of interest, whether of individuals, families, firms, or housing units, changes over time. Individuals enter the population when they are born, immigrate, or attain the age or other status that is used to define the population of interest, and depart when they die, emigrate, move into institutions such as jails, nursing homes, or the military, or in some other way lose that status. Families are ‘born’ when children leave their parents and set up their own independent households or when a divorce or separation breaks up one family into two; they ‘die’ when all members of the original household die or when two households are merged into a single one through marriage or less formal living arrangements. Over a survey’s time span an element may both enter and leave the population, and may do so more than once, as for instance in the case of a person who has several stays in a nursing home during the course of a survey of the noninstitutionalized population. Comparable changes through construction and demolition, or incorporation, merger, and bankruptcy apply to populations of dwellings and firms.

All surveys over time must address the issues raised by these changes in population composition (Duncan and Kalton 1987). In a repeated survey the cross-sectional parameters of objective (a) relate to the population as it exists at the time the sample is drawn for the particular round of the survey. This procedure reflects both the changing values of the variables under study and the changing population over the time period involved. The estimates are thus average values for a form of average population over time.

Panel studies have adopted a variety of strategies for coping with changing populations. The simplest are panel studies of birth cohorts of individuals in the population, where as many members as possible of the original sample are followed throughout the panel period. In effect, these panels ignore all problems of population composition change except death. Disregarding nonresponse and immigration, these studies represent their respective cohorts as they age, while they gradually cease to represent individuals in the original age range. The National Longitudinal Survey cohorts of Older Men and Mature Women are examples. As discussed above, multiple-cohort panel designs avoid this problem by including in subsequent interviewing waves a sample of new entrants into the population.

Panel designs such as that of the PSID, the British Household Panel Survey, and the European Community Household Panel surveys, contain a mechanism for adding to their samples individuals and families that are ‘born’ into the population, so they have the potential for maintaining representative samples of nonimmigrant individuals and families throughout their duration. For example, the PSID began with a probability sample of dwellings that, in turn, provides a representative sample of subunits within those dwellings: households, families, subfamilies, transfer program recipiency units, and individuals. The selection probability of each of these units is identical to the selection probability of the dwelling itself. Other panels composed these subsamples with known selection probabilities through other means. With a properly specified set of rules regarding the definition of units, newly formed subunits of interest (including individuals) enter into the sample with known selection probabilities and reflect corresponding changes that are taking place in the population at large, for example, Hill (1992).

Because they originate from existing panel households, newly formed families and individuals are more ‘clustered’ by family lineage in these designs than in repeated cross-sectional samples; however, the statistical inefficiency of this clustering is heavily outweighed by the analytic advantages of being able to link these newly formed units with information on their families of origin. Thus, information on children setting up independent households can be matched to reliably reported information on their parents, information on ex-spouses can be linked together if the marital split occurred during the panel period, etc. Immigration during the panel period is usually ignored although can be a serious problem for a long-term panel survey confined to a city or a local area, but it is a less serious problem for a national panel survey.

4. Examples Of What Panel Surveys Have Found

To help convey the analytic advantages of panel surveys, we list in this section some prominent examples of panel-survey based analyses.

4.1 Poverty And Welfare Dynamics

Trends in annual US poverty rates (i.e., the fraction of the population with family cash incomes below a set of ‘poverty thresholds’ that vary with family size) are tracked through cross-sectional estimates based on the Current Population Survey. Net change from one year to the next in these rates rarely amount to as much as a single percentage point (US Bureau of the Census 1999), producing perceptions that ‘the poor’ are an ever-present and little-changing group.

By applying event history methods to the time series of annual poverty measurements for PSID sample members, Bane and Ellwood (1986) characterized the nature and determinants of poverty experiences by the length of their ‘spells’ (i.e., continuous periods of poverty). They find that a clear majority (60 percent) of families who first begin a poverty experience do so for short (i.e., spells of 1 or 2 years) periods of time. On the other hand, a subset of poor families (14 percent) have longer-run (8 or more years) spells, and the remaining one-quarter (26 percent) have mediumlength (3–7 years) spells. Thus, there is no single ‘correct’ characterization of poverty—transitory or persistent—since poverty experiences are a mixture of transitory and long-term.

By combining individual spells into simulated lifetime histories, Bane and Ellwood (1994) provide estimates of the likely total number of years of receipt for families just starting to receive cash assistance from the US Aid to Families With Dependent Children program. They find a roughly even distribution of first-time welfare recipients across the three time intervals; roughly one-third (36 percent) have very short welfare experiences, a third (35 percent) mediumlength experiences, and the final third (29 percent) long-term receipt. With welfare, as with poverty, heterogeneity is a key feature, with many families using AFDC as short-term insurance, but a substantial fraction of recipients receiving long-term assistance.

4.2 Intergenerational Transmission Of Economic Status

Correlations in earning between fathers and sons is sometimes treated as a useful measure of intergenerational (im)mobility. Solon (1992) argues that there are two important sources of bias in attempts to estimate such correlations: single year proxies for long-run economic status contain transitory fluctuations that, under plausible assumptions, impart a downward bias to the correlation estimates, and samples used in previous studies (e.g., Mormon brothers, white twins who served in the armed forces) tend to be peculiarly homogeneous, also resulting under plausible assumptions in a downward bias to estimates based on homogeneous than population-based samples.

Data from both the NLSY (in Zimmerman 1992) and PSID (in Solon 1992) have been used to gauge the likely biases. For example, Solon et al. (1992) find that increasing the measurement interval for economic status from one to several years causes their estimate of the father-son earning correlation to jump from 0.25, a number consistent with past studies based on single-year earnings measurement and homogeneous samples, to 0.45. Thus, intergenerational economic mobility appears to be substantially lower than had been depicted in studies based on homogeneous samples and single-year measures of income or earnings.

The more specialized panel studies of Elder and Furstenberg provide much more detail on the process of intergenerational mobility. Using 30-year follow-up data on the Oakland Growth Study cohorts, Elder (1999) found in these children of the Great Depression strong associations among economic hardship, parental psychological well-being, and adolescent wellbeing in intact families.

Fathers who experienced job loss and economic deprivation were more distressed psychologically and prone to explosive, rejecting, and punitive parenting. Preschool-age children in these families, especially boys, were more likely to exhibit problem behaviors, while adolescent girls were more likely to have lower feelings of self-adequacy and to be less goal-oriented. Elder (1999) speculated that the gender and age differences reflected different experiences in families during the deprived times. Adolescent boys reacted to family economic hardship by looking for economic opportunities outside of the home. This time spent taking on responsibility in extra-family contexts reduced the time they spent with their families and may have provided a buffer to the effects of negative family interactions. Younger children and girls were unable to seek such extra-familial opportunities as readily and therefore did not have access to buffering opportunities.

Furstenberg et al. (1987) take a detailed look at intra-and intergenerational mobility in a panel study of a poor teen mothers participating in a program in a Baltimore hospital. Their 20-year follow-up revealed considerably diversity in the mothers’ economic and demographic statuses, with most self-supporting but very few stably married. Most (60 percent) of the daughters of the teen mothers had not themselves become teen mothers and three-quarters enjoyed at least a modest degree of economic success. In contrast, fewer than half of the sons of the teen mothers had attained successful economic trajectories.

4.3 Unemployment And Labor Turnover

Both unemployment and job turnover are most naturally thought about in a ‘spells’ framework: how long does a spell of unemployment last before employment recommences? How long does a spell of employment with one firm last before the worker leaves to take (or look for) alternative employment? Given the inherently longitudinal nature of these processes, it is not surprising that panel surveys have played an important role in research on these topics. While much research on this topic has used administrative records, a panel survey has the advantages of including those who do not receive unemployment benefits, for comparison to recipients, and it continues observing spells after benefits have been exhausted (i.e., it measures total spells, not just insured spells), so one can observe changes in job-finding rates after benefits are exhausted.

An example is Katz and Meyer’s (1990) study of the impact of unemployment insurance on unemployment duration. Katz and Meyer find that probability of recall declines as the duration of a spell increases, for both recipients and nonrecipients. New-job rates, however, are upward-sloping for recipients but relatively flat for nonrecipients. Moreover, there are spikes in both the recall and new-job hazard rates around 26 and 39 weeks (when benefits were typically exhausted in their sample period) for recipients but not for nonrecipients. Thus, not only is there evidence that the approaching end of benefits conditions recipients’ job finding, but it also conditions firms’ recall policies.

4.4 Antecedents Of Life Events

There is great interest in understanding the consequences of important life events such as divorce, widowhood, unemployment, and retirement. Cross-sectional data provide comparisons of, for example, the living standards and labor supply of divorced women and ‘otherwise similar’ married women, or the health of unemployed or widowed men with ‘otherwise similar’ employed or married men. But panel surveys provide valuable post and, interestingly, pre-event data on the same individuals.

For example, Johnson and Skinner (1986) use data from the PSID to examine work hours of women between seven years before and four years after divorce or separation. Consistent with other research, they find that the average work hours of these women rose by half (from 1024 to 1551 hours) between one year before and four years after the divorce or separation. But Johnson and Skinner (1986) also found a big jump (from 744 to 1024 hours) over the six-year period prior to the divorce or separation.

Zick and Smith (1991) analyze the economic antecedents and consequences of widowhood by comparing family-income trajectories of a control group of intact couples with widowed women and men between five years before and five years after the spouse’s death. They too find differences prior to the event, with income changes over the five years prior to the death accounting for about one-quarter of the control widow difference at t 5 and for fully half of the control-widower difference. A clear message from both of these analyses is that impending life events such as divorce and death produce measurable economic and social changes long before the events actually occur. An implication is that fairly long panels are needed to capture the complete dynamic.

Bibliography:

1. Allison P 1982 Discrete-time methods for the analysis of event histories. In: Leinhardt S (ed.) Sociological Methodology 1982. Jossey-Bass, San Francisco, pp. 61–98
2. Atkins E, Cherry N, Douglas J W B, Kieman K E, Wadsworth M E J 1981 The 1946 British birth cohort: an account of the origins, progress, and results of the National Survey of Health and Development. In: Mednick S A, Baert A E (eds.) Prospective Longitudinal Research. Oxford University Press, Oxford, UK, pp. 5–30
3. Bailar B A 1975 The effects of rotation group bias on estimates from panel surveys. Journal of the American Statistical Association 70: 23–30
4. Bailar B A 1979 Rotation sample biases and their effects on estimates of change. Bulletin of the Statistics Institution 48(2): 385–407
5. Bane M J, Ellwood D T 1986 Slipping in and out of poverty: the dynamics of spells. Journal of Human Resources 21: 1–23
6. Bane M J, Ellwood D T 1994 Welfare Realities. Harvard University Press, Cambridge, MA
7. Blau D 1999 The effect of income on child development. Review of Economics and Statistics 81(2): 261–76
8. Bound J, Brown C, Mathiowetz N 2000 Measurement error in survey data. In: Heckman J, Leamer E (eds.) Handbook of Econometrics. North Holland, Amsterdam, Vol. 5
9. Call V R A, Otto L B, Spenner K I 1982 Tracking Respondents: A Multi-Method Approach. Lexington Books, Lexington, MA
10. Cannell C F, Miller P V, Oksenberg L 1981 Research on interviewing techniques. In: Leinhardt S (ed.) Sociological Methodology, 1981. Jossey-Bass, San Francisco pp. 389–437
11. Clarridge B R, Sheehy L L, Hauser T S 1978 Tracing members of a panel: A 17-year follow-up. In: Schuessler K F (ed.)
12. Sociological Methodology, 1978. Jossey-Bass, San Francisco, pp. 185–203
13. Davis J, Smith T W 1994 General Social Surveys, 1972–1994: Cumulative Codebook. National Opinion Research Center, Chicago
14. Deutsches Institut fur Wirtschaftsforchung 2000 German SocioEconomic Panel Home Page, http://www.diw-berlin.de/soep/
15. Duncan G, Kalton G 1987 Issues of design and analysis of surveys across time. International Statistical Review 55: 97–117
16. Elder G 1999 Children of the Great Depression: Social Change in Life Experience, 25th Anniversary Edition. Westview, Boulder, CO
17. Eizenman D R, Nesselroade J R, Featherman D L, Rowe J W 1997 Intra-individual variability in perceived control in an older sample: The MacArthur successful aging studies. Psychology and Aging 12: 489–502
18. Freedman D, Thornton A, Camburn D 1980 Maintaining response rates in longitudinal studies. Sociological Methods and Research 9: 87–9
19. Furstenberg Jr F F, Brooks-Gunn J, Morgan S P 1987 Adolescent Mothers in Later Life. Cambridge University Press, New York
20. Ghangurde P D 1982 Rotation group bias in the LFS estimates. Survey Methodology 8: 86–101
21. Hill M S 1992 The Panel Study of Income Dynamics. Sage, Newbury Park, CA
22. Institute for Economic and Social Research, University of Essex 2000 British Household Panel Survey Home Page, http://www.irc.essex.ac.uk/bhps
23. Institute of Education, Center for Longitudinal Studies 2000 British National Child Development Study Home Page, http://www.cls.ioe.ac.uk/Ncds/nhome.htm
24. Johnson W, Skinner J 1986 Labor supply and marital separation. American Economic Review 76(3): 455–69
25. Kalton G 1986 Handling wave nonresponse in panel surveys. Journal of Official Statistics 2: 303–14
26. Katz L, Meyer B 1990 The impact of potential duration of unemployment benefits on the duration of unemployment. Journal of Public Economics 41(1): 45–72
27. Kish L 1986 Timing of surveys for public policy. Australian Journal of Statistics 28: 1–12
28. Lepkowski J 1989 The treatment of wave nonresponse in panel surveys. In: Kasprzyk D, Duncan G, Kalton G, Singh M (eds.) Panel Survey Design and Analysis. Wiley, New York pp. 348–74
29. MaCurdy T, Mroz T, Gritz R M 1998 An evaluation of the National Longitudinal Survey on Youth. Journal of Human Resources 33(2): 345–436
30. Mathiowetz N Duncan G 1984 Temporal patterns of response errors in retrospective reports of unemployment and occupation. In: Proc. Survey Res. Meth. Sect. American Statistical Association, Washington, DC, pp. 652–7
31. Monitoring the Future 2000 Monitoring the Future Home Page, http:/monitoringthefuture.org/
32. Moss L, Goldstein H 1979 The Recall Method in Social Surveys. University of London Institute of Education, London
33. Solon G 1992 Intergenerational income mobility in the United States. American Economic Review 82(3): 393–408
34. Smith T W 1994 Is there real opinion change? International Journal of Public Opinion Research 6: 187–203
35. Statistics Canada 2000 Guide to the Labour Force Survey Statistics Canada web site, http://www.statcan.ca/
36. Traugott M, Katosh K 1979 Response validity in surveys of voting behavior. Public Opinion Quarterly 79: 359–77
37. Tuma N B, Hannan M T 1984 Social Dynamics: Models and Methods. Academic Press, Orlando, FL
38. US Bureau of the Census 1978 The Current Population Survey: Design and Methodology. Technical Paper 40. US Government Printing Office, Washington, DC
39. US Bureau of the Census 1999 Poverty in the United States: 1998 (P60–207). US Government Printing Office, Washington, DC
40. US Bureau of the Census 2000 Survey of Income and Program Participation Home Page, http://www.sipp.census.gov/sipp/sipphome.htm
41. US Bureau of Labor Statistics 2000a Current Population Survey Home Page, http://www.bls.census.gov/cps/cpsmain.htm
42. US Bureau of Labor Statistics 2000b National Longitudinal Surveys Home Page, http://stats.bls.gov/nlshome.htm
43. Yamaguchi K 1991 Event History Analysis, Applied Social Research Methods, Vol. 28. Sage, Newbury Park, CA
44. Zick C, Smith K 1991 Patterns of economic change surrounding the death of a spouse. Journal of Gerontology 46(6): S310–20
45. Zimmerman D J 1992 Regression toward mediocrity in economic stature, American Economic Review 82(3): 409–29.