The reliability of recall measurement in assessing migrant reintegration: Evidence from Ethiopia, Somalia, and Sudan

This article asks whether and under what conditions the use of retrospective or “recall” measurement is reliable. Migration researchers are often forced to retrospectively measure baselines when evaluating the impact of inter - ventions due to the transitory nature of migration, de - veloping country contexts, and hastily assembled policy programmes, a situation exacerbated by Covid-19. This ar - ticle first theoretically considers the extent to which this approach is reliable and likely to result in biased estimates, as well as its broader advantages, disadvantages, and rec - ommendations for best practice. It then considers the case of the “IMPACT” evaluation of the EU-IOM Joint Initiative for Migrant Protection and Reintegration in which 1774 Ethiopian, Somalia, and Sudanese migrant returnees in 2021 were assessed on a range of reintegration measures, 1095 of whom were measured retrospectively. Regression analyses demonstrate that those measured retrospec - tively give more negative scores on several “Reintegration Sustainability Scores” in line with some theoretical expec - tations but contrary to others. However, this— mostly non-statistically significant— effect is largely diminished when the small minority who report finding it difficult to remem - ber the baseline period are removed— suggesting that any retrospective measurement effect results from memory


INTRODUC TI ON
This article asks to what extent retrospective survey measurement is reliable and to what extent, and under what conditions, it can be optimal. Retrospective survey measurement, also known as "retrospective enumeration", is characterised by asking individuals about their past-for example, "thinking about your life two years ago, how happy would you say you were on a scale of 0-10?". These can be used to produce baseline estimates in the calculation of over-time change. This article uses the case of a survey of returnee migrants in Ethiopia, Somalia, and Sudan to investigate retrospective survey measurement. This survey forms one part of the European Union-

International Organisation for Migration Joint Initiative for Migrant Protection and Reintegration, launched in
December 2016 with €350 million funding from the EU Emergency Trust Fund for Africa (EUTF) to reintegrate migrants in their countries of origin. The programme is expected to contribute to International Organisation for Migration's (IOM) Integrated Approach to Reintegration by addressing returnees' economic, social, and psychosocial needs and to include communities of return in reintegration planning and support.
Scientifically, understanding the extent to which retrospective or recall questions can be used to measure baseline scores is of interest because of ongoing arguments about the extent to which such measures are subject to biases regarding memory, consistency, survival, and social desirability (see review below, including Hipp et al., 2020;Thigpen, 2019). More profoundly, it raises questions about human ability to understand his or her own past and world accurately and objectively or whether such memories and understandings are function to serve cognitive needs in the present. More broadly, comparing retrospective measurement to other forms of baseline measurement-i.e. contemporaneous (also known as prospective) baseline measurement as part of panel dataallows us to consider under what conditions each of the approaches represents the best or "least bad" available approach.
The extent to which retrospective measurement is reliable is also substantively important. Migration, specifically, is likely to remain one of the world's most important and complex political challenges throughout the 21st century, with vast economic and humanitarian consequences and raising profound legal-and rights-based questions for millions of people worldwide. Understanding the actual effects of interventions aimed at improving reintegration outcomes is thus of overwhelming practical importance for advocacy organisations, governments, communicators, policymakers, and those working in politics who want to know what constitutes a sustainable, effective, and valuefor-money policy. There are already significant scientific findings on the challenges and determinants of successful reintegration (Arowolo, 2000;Fentaw, 2018;Mercier et al., 2016) and assisted voluntary return and reintegration more generally (Koser & Kuschminder, 2015;Kuschminder, 2022;following Covid-19, Le Coz & Newland, 2021). However, the transitory and international nature of migration, often involving developing countries, means that traditional panel and longitudinal approaches, particularly over lengthy periods, are not always possible. Therefore, identifying if and under what conditions retrospective measurement can be used has plenty of potential uses for policy and programme impact evaluation, regarding the Joint Initiative for Migrant Protections and Reintegration and far beyond, particularly following the disruptive impact of the Covid-19 pandemic. Moreover, testing the reliability of retrospective measurement in the field of migration-as has recently been done in a wide range of other policy areas-may bring to the fore broader and more nuanced lessons for practitioners across disciplines. bias rather than, for example, consistency bias. No evidence is found to support several theoretically derived interaction effects. Determinants of self-reported memory are demonstrated and recommendations for usage of retrospective measurement are provided, based on these findings.

RECALL MEASUREMENT IN ASSESSING REINTEGRATION
The article proceeds as follows. First, the literature on retrospective survey measures is considered, including previous findings on its potential shortcomings, benefits, and best practices. Second, several hypotheses are offered to explain potential discrepancies between retrospective and contemporaneous measurement. Third, the methods and data, including the specific case and survey experiment, are outlined. Fourth, several analyses of the effects of retrospective versus contemporaneous measurement on a range of reintegration indicators, as well as theoretically relevant interaction effects, are presented. Finally, broader ramifications and recommendations based on the analyses are discussed, as well as its shortcomings and next steps for research.

Retrospective survey measurement
Increasingly transnational, fast-moving, and unpredictable crises and transformations-recently including pandemics, migration crises, climate-related disasters, digitalisation and internet threats, terrorism and unconventional warfare, and economic and financial crises-have often left policymakers at various levels of governance scrambling to devise and execute post-facto solutions. Resultant policy programmes-some of which have budgets in the billions of dollars and cross numerous national boundaries-are therefore often hastily assembled. One example of the challenges posed by the measurement and evaluation of these large and reactive programmes is the need to measure change in individuals, ideally requiring pre-(also known as baseline or ex ante) and post-(also known as topline or ex post) programme measurement despite no measurement being taken before the programme began. Moreover, even using genuine baseline data to measure change presents numerous problems and biases for researchers, such as panel response, panel fatigue, and panel conditioning and the "Hawthorne effect", with one result being the need to top-up samples to maintain representativeness (Scott & Alwin, 1998: 16).
Researchers have sought to bypass this issue by retrospectively measuring the baseline; in survey research, notably, by asking individuals to recall their situation ex-ante as well as describe their ex-post situation at the time of responding to the survey (Hipp et al., 2020). This approach has been recently utilised in assessing the effects of the Covid-19 pandemic and its various public policy responses (Giorgio et al., 2020;Hipp et al., 2020;Li et al., 2020;Ran et al., 2020;SORA, 2020). Researchers investigating long-term socio-economic changes for which there is no available panel data have also turned to retrospective questioning, the data from which is used to produce panel data post facto (e.g. Boucher et al., 2007;Fleisher & Wang, 2005;Zhou, 2000). Moreover, researchers investigating developing countries or transitory groups, such as migrants, have needed to make use of retrospective questions, given the difficulty in tracking individuals over time in such contexts (Gibson & Kim, 2010).
Although retrospective measurement has obvious benefits for both researchers and policymakers, it has also been argued to result in notable sources of measurements error, resulting in less accurate responses than contemporaneous measurement (Coughlin, 1990;Hipp et al., 2020;Solga, 2001). First, retrospective questions can only be asked to survivors, both in the literal sense and the broader sense of those that are available for questioning (Scott & Alwin, 1998; although this is also the case for panel data). Second, retrospective questions place high cognitive demands on respondents (Durand et al., 2015;Himmelweit et al., 1978;Yan & Tourangeau, 2007) notably in terms of remembering topics that are or were unimportant to them (Belli, 1998;Bound et al., 2001;Coughlin, 1990;Kennickell & Starr-McCluer, 1997;Peters, 1988;Pina Sánchez et al., 2014;Smith & Thomas, 2003;Teitler et al., 2006) or being unable to accurately date life events, reducing our confidence in temporal ordering (Jacobs, 2002;Paull, 2002) or negatively affecting the accuracy of more volatile events (Gibson & Kim, 2010). Third, individuals have been argued to give biased responses to avoid cognitive dissonance with their analogous contemporaneous self-assessments, narratives, or self-understandings (Barsky, 2002;Jaspers et al., 2009;Schmier & Halpern, 2004;Teitler et al., 2006;Yarrow et al., 1970), or to maintain consistency with contemporaneous social norms and values (a form of social desirability bias; Coughlin, 1990;Himmelweit et al., 1978).
Both of the latter constitute a form of consistency bias that would overestimate similarity between the retrospective baseline measurement and the contemporaneous "endline" (or ex-post or post-treatment) measurement (Cacioppo, 2002: 130-132;Leising, 2011;Shachar & Eckstein, 2007). Aside from this potential for underestimating difference, several biases would also exaggerate the positivity with which respondents retrospectively assess their past, notably "rosy retrospection" (Mitchell et al., 1997;Mitchell & Thompson, 1994;Zurbriggen et al., 2021), the related "euphoric recall", and egocentric bias, whereby, amongst other things, individuals overestimate their own outcomes in hindsight (see also "hindsight bias"; Connolly & Bukszar, 1990;Roese & Vohs, 2012). Alternatively, however, there may be a negative effect of retrospective measurement if the treatment leads individuals to reappraise ex post their subjective assessment of their situation ex ante, i.e. their standards are changed by the treatment so that they judge their previous situation more negatively than they would have at the time, a form of systematic measurement error whereby the treatment affects the baseline (Cochran, 1968;Saris & Revilla, 2015).
However, empirical evidence remains too mixed to entirely rule out retrospective measurement, certainly in comparison to other approaches. Thigpen (2019: 453) finds that 'the passage of time has a weak, uncertain influence on recollection' and concludes that 'the recalled answers display a sufficient degree of correspondence to prospectively collected (i.e. contemporaneously collected) response to have faith in analyses using retrospective survey data regarding traits within the past five years or less.' Similarly, Peters (1988) finds substantive agreement between retrospective and panel measurement approaches, with errors found to increase with more cognitively demanding questions, and Pierret (2001: 439) finds that "coefficients … did not change greatly" between the two approaches. Smith and Thomas (2003: 47) compare panel and retrospective approaches for measuring migration histories, concluding that retrospective questions 'can elicit useful information if we can isolate those events which people remember from those which they forget', which they suggest can be more easily done by using retrospective on more recent events (with the last 2 year years) of longer duration (over 6 months), linking questions to other high salience life events (e.g. marriage, having children), asking multiple members of the same household the same retrospective questions, and having interviewers record the quality of the interview. Overall, they suggest that 'the current presumption against the use of long-term recall questions in field surveys ignores a potentially rich source of data'. Müggenburg (2021) also makes several suggestions on the content of questions and survey design that are likely to reduce bias in retrospective measurement, with theoretical justifications for each. None of these studies validate data from either method so that the accuracy of retrospective data remains a source of 'continued debate' (Gibson & Kim, 2010: 687). Overall, Scott and Alwin (1998: 16, 23; see also Müggenburg 2021, for similar conclusion) note that both retrospective and prospective approaches contain several potential sources of bias so that 'if both prospective and retrospective data have merits, then the ideal may point to a middle ground solution' so that 'retrospective measurement is, for most social researchers, an indispensable tool'.

Hypotheses
Hypothesis 1. Retrospective measurement of reintegration has a positive effect on baseline reintegration metrics compared to contemporaneous measurement (consistency and rosy retrospective biases).
Hypothesis 2. Retrospective measurement of reintegration has a negative effect on baseline reintegration metrics compared to contemporaneous measurement (systematic measurement error).
Hypothesis 3. Selfassessed memory of the time that the baseline refers negatively affects the effect of retrospective measurement on reintegration metrics (memory bias).

The IMPACT study, methods, and data
To test the above hypotheses, this article uses data from the "IMPACT" study, which evaluated the EU-IOM Joint Initiative for Migrant Protection and Reintegration (henceforth, the EU-IOM Joint Initiative). The EU-IOM Joint Initiative comes in the context of a change in thinking around migrant re-integration. Whereas re-integration was formerly conceived of-or at least acted upon-in terms of cash for returnees, larger and more developed programmes since the mid-2010s increasing conceive of reintegration programmes in structural terms involving governments and communities, in addition to individuals. Similarly, the stated objectives of such programmes, and thus the ways in which their impacts are measured, have turned away from purely economic indicators to also include social and psychosocial measured. Overall, practically, this has meant that these more sophisticated and ambitious reintegration programmes have left their administrators, such as the IOM, responsible for delivering a broad range of services to returning migrants to assist their reintegration across numerous aspects of life, including business start-up support, training, education, and job placement, financial support and management, improved access to services, and psychosocial support including community integration, support groups, and mentorship.
The EU-IOM Joint Initiative was launched in December 2016 with €350 million in funding from the EU Emergency Trust Fund for Africa (EUTF) and claims to be 'the first comprehensive programme to save lives, protect and assist migrants along key migration routes in Africa'. 1 It does so by assisting migrants who decide to return to their countries of origin through cooperation between partnering state and non-state actors that aims to support both migrants and their communities across 26 African countries. The stated objectives of the Joint Initiative include providing direct assistance and enabling the assisted voluntary return of migrants stranded along the migration routes and supporting the reintegration process of returning migrants in an integrated approach which addresses economic, social, and psychosocial dimensions and fosters the inclusion of communities of return, amongst other objectives related to outbound migrants.
Overall, the programme is considered by the IOM to be a flagship and first attempt to apply an "integrated approach to reintegration" in contrast with 'traditional' return and reintegration programmes (for detailed information on what the programme does and the IOM's reintegration approach, see IOM, 2019). However, the programme was also implemented in a rush under strong political pressure (Dennison & Geddes, 2019).
To measure the EU-IOM Joint Initiative, and reintegration programmes more generally, the IOM's 2017 "Measure" project defined a framework for the measurement of reintegration outcomes of returning migrants: Reintegration Sustainability Score (RSS) and a methodology to compute the RSS with the data gathered through the survey questionnaire (Samuel Hall and IOM, 2017). Guidelines on the usage of this measurement framework recommend the administration of the RSS twice: the first time to establish a 'baseline' measurement-around 3 months after return-and a second time to establish an 'endline' measurement-at least 1 year after return.
Building on this work, the IMPACT study, starting in 2019, sought to assess the impact of the reintegration assistance provided by the EU-IOM Joint Initiative, although restricted to Ethiopia, Somalia, and Sudan. It is the data from that study that this article relies upon (see methodological report, Eager et al., 2020). The IMPACT study used a quasi-experimental design whereby reintegration outcomes of members of a treatment group composed of programme beneficiaries are systematically compared with the (re)integration outcomes of members of a "calibration" (i.e. control) group. This latter group is composed of non-migrant individuals who were not assisted by the programme and are matched to one member of the treatment group based on a defined set of sociodemographical criteria including residing in the same communities where the beneficiaries live. The RSS in this study was modified through a comprehensive review process, which resulted in an enhanced tool called RSS+. The RSS+ includes a set of 'retrospective' questions (not included in RSS) which aim at allowing the computation of a 'retrospective-baseline' RSS score in case baseline data is not available for the respondent interviewed.
The RSS+ study is an evolution of the original RSS questionnaire and, crucially for this article, contains retrospective questions as well as additional information on the respondent's understanding of retrospective questions. The introduction of retrospective-baseline questions into RSS+ was motivated by both practical and methodological considerations. On the practical side, the outset of the pandemic caused a significant decrease in the influx of new returns, making it therefore impossible to achieve the targeted sample size with new returns alone, given that baseline data is available only for a small fraction of the beneficiaries returned before the launch of the IMPACT study. On the methodological side, provided proper design and methodological precautions are in place, it might be argued that retrospective enumeration may present possible benefits in terms of bias reduction and elaboration of the questions by the respondents.

Data
Data continues to be collected for the IMPACT study at the time of writing, including a full control group that did not receive IOM reintegration assistance. Such data will continue to be collected until around the end of 2023.
However, the 1774 observations collected by the IMPACT Study by early 2021 can already be used to test the above hypotheses. Although this survey included many dozen variables, for the purposes of this article, only ten independent variables and four dependent variables are utilised (the comprehensive Reintegration Sustainability Score and its three constituent parts: economic, psychosocial, and social). The variable names, their possible responses, and the distribution of those responses for the purposes of this study are shown in Table 1. It is worth noting that the large, complex nature of RSS data and the migration context often results in missing variables, attrition, etc., although in this case the regression models of RSS scores contain 1648 observations, a relatively minor drop from the full sample. Most of these variables and the justification for their inclusion are relatively selfexplanatory. Psychological problems records responses to the question "Do you often suffer from any of the following? Feeling angry; Feeling sad; Feeling afraid; Feeling stressed; Feeling lonely; Feeling low self-worth; Difficulty concentrating" and is included as a control variable in case such symptoms lead to memory problems. Memory records responses to the question "When asked about things related to the past (3 months after return), how easy or difficult was it to remember your situation back then and answer the questions?" and is included to test whether such responses affect the effect of the form of baseline measurement (i.e. whether they impair the accuracy of retrospective measurement). Because this is only asked to those individuals responding retrospectively, it is not included in the main regression analyses.

Methods
Hypothesis 1 and 2 are tested by comparing baseline scores between those who were and were not retrospectively measured in two ways. Most naively, this can be done by simply comparing the four mean baseline RSS scores amongst those measured contemporaneously and those measured retrospectively. However, there  Mercier et al., 2016), and measurement issues such as time between the baseline and its measurement, whether the measurement was done face-toface or via phone (Holbrook et al., 2003), and whether it is done with the migrant or member of their family or cohabitee (Davin et al., 2019). Propensity Score Matching (PSM) is not used given the strong recent criticism of this approach when-as in this case-one is testing the effects of a non-randomly distributed treatment; namely that it 'accomplishes the opposite of its intended goal -thus increasing imbalance, inefficiency, model dependence, and bias' (King & Nielsen, 2019: 435) and 'there is no way to know in advance whether the method will work' (Peikes et al., 2008: 62; for further discussion on PSM criticism, debates, and under what conditions it may be used see Guo et al., 2020;Smith & Todd, 2001;Wang, 2021).
Importantly, however, given that several methodological sources of bias stemming from retrospective measurement are related to memory, it is also important to see if our results vary by one's self reported ease of memory of the time period in which the baseline was taken (Hypothesis 3). This is done simply by repeating the four regression analyses (one for each of our RSS measures) with those not reporting that remembering that time was easy (21 per cent of retrospective responders) being excluded from the analysis. Finally, Hypothesis 4a and 4b can be tested by running the same regression analyses, but including interaction effects for level of education and for the duration of time between the baseline measurement and the time period to which it refers (i.e. 3 months after return).

Analyses
The simple mean RSS scores of those whose baseline was measured contemporaneously and those whose baseline was measured retrospectively are shown in Table 2. We can clearly see only minor differences between the two groups across all four of the RSS scores. Overall, these initial averages provide most support for the null hypothesis. This is further investigated by considering variation in the above responses according to days since measurement, with the latter divided into five groups for ease of comparison in Table 3 below. Overall, we see no evidence of systematic variation, either positively or negatively, according to days since measurement.
In Table 4, our regression analyses are presented. Model 1 shows the predictors of the comprehensive RSS score, on which retrospective measurement has a non-statistically significant negative effect of 0.032 (the full scale is between 0 and 1), an effect which is larger than, for example, education. This finding offers some support for Hypothesis 2, although without statistical significance we cannot be certain about the finding in the population. However, notably, as shown in model 2, the effect of retrospective measurement falls greatly when the 20 per cent of those measured retrospectively who reported finding it not easy to remember the time-period are removed from the analysis. This suggests that the negative effect observed in model 1 is partially a result of importantly and shown in models 4, 6, and 8, this effect becomes considerably smaller when those who report not finding it easy to remember are removed and, in the case of the effects of the economic and social RSSs, becomes non-statistically significant in all cases. Some of the control variables also display consistent and theoretically interesting effect. When the primary respondent to the survey is the returning migrant her-or himself, the RSS score increases in every model, except the two relating to social services (perhaps reflecting the community nature of these indicators). The use of phone interview rather than in person increases the RSS score in every case, while there is considerable evidence that a higher age has the same effect. Overall, we find the strongest evidence to support the null hypothesis, which in the case of the comprehensive RSS score becomes stronger when those who cannot remember are removed.
Moving on to our test of Hypothesis 4a or 4b displayed in Table 5, we can see that neither days since the baseline (models 1 and 2) nor education level (models 3 and 4) have either large or statistically significant effects on the effect of retrospective measurement on the comprehensive RSS. As such, we find little evidence to support Hypothesis 4a or 4b so, theoretically, memory bias does not seem to be a function of lack of education or time since the event. Models 5 and 6 also show how the effect of retrospective measurement varies by country. We can see that the negative effect of retrospective measurement is exacerbated by being in Somalia rather than Ethiopia and reduced by being in Sudan rather than Ethiopia. Notably, it is only when this interaction is included do we see a statistically significant main effect of retrospective measurement on the comprehensive RSS.

Determinants of retrospective memory
Given the clear importance of self-reported retrospective memory in determining the effect of retrospective measurement on one's RSS score, it is worth considering further the determinants of self-reported retrospective memory. To do so, a dichotomous variable of retrospective memory is produced (0 for those saying it is easy to remember 3 months after return and 1 for those either stating that it is difficult to do so or neither difficult nor easy). Along with the same socio-demographic and measurement control variables as the earlier regressions, the resultant regression analysis predicting ease of memory is shown in Table 6. Notably, none of gender, education, age, or days since the baseline measurement influence one's memory. Instead, we see that being in Sudan, rather than Ethiopia, decreases the chance that one says it is not easy to remember whereas, unsurprisingly, having psychological problems increases the chance. Similarly, doing the interview over the

TA B L E 4
Regression analyses of determinants of RSS scores, by inclusion and exclusion of those who self-report as not finding it easy to remember the baseline time period. (1) (3)    Note: Standard errors in parentheses and clustered by country. ***p < 0.01; **p < 0.05; *p < 0.1.

TA B L E 4 (Continued)
TA B L E 5 Regression models testing interaction effects between retrospective measurement and days since baseline, education, and country. (1) (3)

TA B L E 5 (Continued)
phone, rather than face-to-face also increases one's chance of saying that they cannot remember the period easily.

Discussion and recommendations
This article considered to what extent retrospective survey measurement is reliable and what are its advantages, disadvantages, and best practices. The transitory nature of migration and the developing country contexts for-at times, hastily assembled-policy programmes often make panel and randomised control trial designs impractical when evaluating migration reintegration projects, a situation recently exacerbated by Covid-19. Migration researchers therefore often turn to retrospective baseline measures, an approach that theoretically may result in measurement biases in TA B L E 6 Determinants of not finding it easy to remember the time of 3 months after return.
(1) Regression analyses demonstrate that those measured retrospectively give more negative scores on several "Reintegration Sustainability Scores"-though these effects were mostly not statistically significant-in line with theoretical expectations regarding measurement error, but contrary to several theorised biases. However, this effect weakened considerably when the small minority who report finding it difficult to remember the baseline period are removedsuggesting that any retrospective measurement effect partially results from memory bias rather than, for example, consistency bias. No evidence was found to support several theoretically derived interaction effects.

Not easy to remember
This article therefore makes several contributions. First, it is demonstrated that the effect of giving retrospective than contemporaneous baseline measurements is negative where it exists, contrary to our expectations from at least three well-documented cognitive biases but consistent with the possibility of endogeneity. Second, it is shown that this effect partially disappears when the small minority of retrospective respondents who state that it was not "easy to remember" the time of the baseline are removed. As such it is argued that much of the retrospective bias comes from memory and so, third, it is recommended that researchers thus ensure that a measure is included on self-reported memory when using retrospective measurement, notwithstanding the risks of As such, because retrospective measurement has significant practical and logistical advantages over panel approaches, notably in terms of efficiency, subject as panel approaches are to attrition and other biases that are especially problematic in dealing with transient populations, this article recommends that researchers ideally utilise both forms of baseline measurement in impact evaluations. When using retrospective measurement, however, it is vital to: 1. Simultaneously gather data on self-reported ease-of-memory of the time being measured.
2. Systematically test for variation in the results according to self-reported memory.
3. Because phone interviews are found to increase self-reported problems in memory, it is recommended that face-to-face interviews are especially prioritised.
This article is necessarily limited and represents a first attempt, both theoretically and empirically, to consider the appropriate use of retrospective measurement in migration research. The lack of randomisation-necessary in the case of this programme-between those measured retrospectively and those measured contemporaneously may have biased the results if there were differences between the two groups that co-vary with any factors affecting the reintegration measures-though the, primarily, null findings reduce the likelihood that this was the case. Furthermore, further research should be made into how retrospective measurement accuracy diminishes over time, regardless of self-assessed memory of the period. This study found that people's reported memory is not affected by the time since the period being remembered, contrary to expectations regarding memory bias (although in line with Smith and Thomas's (2003) expectation of accuracy within 2 years since the mean for this study was 602 days with a median of 509). This by no means suggests that memory bias will not increase at some point or in some circumstances though, with the lack of interaction effect perhaps reflecting the relatively short time period, the salience of the time period of the baseline measurement in an individual's life, or biases increasing but "cancelling out" over time (some combination of survivor bias, cognitive demand, cognitive consonance, rosy retrospection, euphoric recall, and egocentric bias, with some of these having theoretically plausible contrary effects).
Future research should utilise the IMPACT study's growing body of data, including the future inclusion of a control group, to produce genuine randomised control trials from which the effect of retrospective measurement can be further tested. Similarly, as the number of individuals increases, it will be possible to effectively test how participation in various reintegration activities (such as starting a microbusiness, done by 410 participants in this initial sample) affected-above all-the RSS scores, but also secondarily the effect of the type of baseline measurement. Furthermore, given the seeming centrality of memory bias to problems of retrospective measurement-and the inability of this article to find socio-demographic determinants of it-it is vital to further investigate the causes of variation in this memory variable.

PEER R E V I E W
The peer review history for this article is available at https://www.webof scien ce.com/api/gatew ay/wos/peerrevie w/10.1111/imig.13177.

DATA AVA I L A B I L I T Y S TAT E M E N T
The full dataset used for this article contains confidential information. However, a reduced version can be provided upon request for replication purposes.