Raising Take-Up of Social Assistance Benefits through a Simple Mailing: Evidence from a French Field Experiment (2024)

1 In2009the regime of means-tested social assistance benefits in Franceunderwent a profound transformation with the introduction of a new benefitcalled the Revenu de Solidarité Active (RSA—the Solidarity Earned IncomeSupplement), which replaced its longstanding predecessor known as theRevenu Minimum d’Insertion (RMI—Minimum Integration Income). With aneye towards strengthening work incentives and avoiding the well-known butsocially and economically harmful “welfare trap”, the new income-supportbenefit was designed to sharpen the incentives for recipients to earn labourmarket income. More specifically, on the margin, increases in labour marketearnings would no longer be clawed back by a commensurate reduction insocial assistance benefit amounts. The aim was to guarantee that starting orreturning to work, as well as increasing the number of hours worked, wouldunambiguously increase the total income received by recipient households.An unexpected, perplexing, and disconcerting effect of this reform, however,was an apparent decrease in the overall take-up rate among seeminglyeligible low-income households. Indeed, a study conducted by the evaluation council of the RSA that assessed the new benefit concluded that68% ofthe newly eligible beneficiaries did not take-up their benefits (Domingo andPucci [2011]). The extent of this phenomenon of what was viewed as undulylow take-up resulted in the development and implementation of yet anothernew reform labelled the “prime d’activité” (activity bonus) in2016.

2 The unduly low take-up of social benefits, which is a major policy concernfor other countries as well as France, implies that social policy measurestargeted at vulnerable socio-economic groups often fail to reach theirintended beneficiaries, which undermines the dual objectives of alleviatingpoverty (i.e., the passive function) and ultimately facilitating integration intothe labour force (i.e., the active function). In addition to these policy challenges, it poses an economic puzzle. The fact that a significant number ofpotential beneficiaries fail to take up benefits or to maintain their eligibilityonce their claim has been approved—especially in the case of passivebenefits—runs directly counter to rational choice-making behavior that isset out in standard neo-classical microeconomic theory.

3 Several competing economic explanations of why eligible individuals orhouseholds fail to take up benefits have been proposed in the economicsliterature. We evoke three channels in no particular order. First, they mightbe instrumentally rational (as in standard, basic microeconomic theory) butnot know that they are eligible because the relevant information is unavailable to them, as suggested in the application of Daponte et al. [1999]. Thisparticular behavioral effect involves little more than the non-transmission ofinformation. The second possible channel is that they are aware of theireligibility, and are instrumentally rational, but upon weighing the perceivedcosts against the perceived advantages, they decline to take up benefits. Thecharacteristics of their preferences, such as the social stigma associated withenrollment (as in applications by Currie [2004] and Moffitt [1983]), might becomplex. A related factor is the existence of transactions costs, as mentioned in the application of Anderson and Meyer [1997] involving the take-upof UI benefits in the USA.

4 In regards to the third channel, potential claimants might know that theyare eligible, and might have standard preferences, but be boundedly rational. In this case, they might be tempted by the benefits yet fail to take themup for reasons such as procrastination or psychological aversion to programcomplexity. The former factor is central to applications involving savingsbehavior, such as those authored by Jones [2010] and Madrian and Shea[2001]. The latter factor is central to the analysis of social assistance regimesby Bertrand et al. [2006]. The relative applicability of these three distinctchannels is an interesting question for applied economic research, but theyare particularly pertinent for public policy analysis because they involvesharply different repercussions.

5 The primary objective of this paper is to evaluate an intervention thatcould increase the take-up of benefits to which one is entitled. Our methodological approach consists of an experimental evaluation in the form of arandomized control trial (RCT). This technique is frequently utilized in therealms of behavioral economics and public policy evaluation. [1]In such applications, the intervention of the experiment sometimes consists of applyingminor changes to the presentation of information and observing subsequentresponses.

6 The outcome variable for our RCT is not the initial choice among theeligible population to apply for RSA benefits, but rather the choice to complywith the program’s key qualifying regulations in order to maintain theireligibility to collect benefits. Given our focus on this particular event, ourresults pertain not to the initial take-up rate, but rather to the continuationrate (which is the flipside of the attrition rate) conditional on initial receipt.Shortly after taking-up RSA benefits, recipients are required to meet with acase-worker, participate in a guidance interview, and sign a contract thatengages him/her to take certain steps—all with an eye towards becomingintegrated or reintegrated into the labour market. A surprising number ofrecipients place their benefit entitlement in jeopardy by failing to complywith these requirements. As we explain below, the goal of the treatment orintervention that we apply is to encourage compliance with this obligation.

7 Enforcement of regulations imposing requirements on beneficiaries hasbeen tightened in France (as well as in other countries) in recent decades. InGermany since2005, the payments to beneficiaries of the minimum incomebenefit (called “Arbeitslosengeld II”) can be reduced in their amount orterminated altogether if they are deemed to have refused a job offer, aninternship, or any activation measure. In order to benefit from the “Jobseeker’s Allowance” in the UK, one is obligated to sign an agreement certifyingwhich steps must be carried out in order to find work. After initiating theclaim, the beneficiary is supported and monitored by a coach. Payment ofthe allowance may be stopped if the contract is not respected. In all threecountries, therefore, changes in the enforcement standards can affect thenumber of continuing beneficiaries of these programs. The topics of failureto comply with eligibility requirements and failure to participate in the counselling services and other related programs have been highlighted in theFrench literature by Dmitrijeva et al. [2015], but to our knowledge, the linkbetween take-up of social assistance benefits and non-compliance withthese regulations has not been treated in the literature.

8 It is thought that one important explanatory variable driving the phenomenon of non-compliance is the quantity and the nature of the informationthat potential beneficiaries have at their disposal. Following the techniqueadopted by Bhargava and Manoli [2015], we employ a protocol that changesin exogenous fashion information regarding the program. Employing thetechnical term attributed to Thaler and Sunstein [2008], we apply two distinct nudges to two separate treatment groups. As defined by these authors,nudges refer to “any aspect of the choice architecture that alters people’sbehavior in a predictable way without forbidding any option or significantlychanging their economic incentives”. According to the survey paper byFrench and Oreopoulos [2017], they often involve minor tweaks to themodes of communication, are often cheap to implement, and focus on rendering decisions more salient and simpler to understand. [2]

9 Our experimental design may give some insight into the applicability ofsome of the three channels that are mentioned above. One of our threeprotocols provides additional information that is germane to the valuation ofboth monetary and non-monetary aspects associated with compliance withthe requirements. This in turn might counteract the effects of perceivedtransactions costs and other sources of reticence, thus altering the trade-offbetween costs and benefits of the decision to participate in the interview. [3]Such an intervention has been found to generate important effects in otherwell-known experiments, such as those conducted by Chetty and Saez[2013] and Saez [2009]. [4]Our other protocol involves only a simplification ofthe information provided to the beneficiaries and might apply in cases ofnon-participation due to complexity aversion or to limited attention.

10 The RCT was carried out in the French département of Seine-et-Marne,which is located within the metropolitan Paris area southeast of Paris proper,and accounts for about11% of the population of the greater Paris area. Thisgeographical and administrative unit is the French counterpart of a populouscounty. Seine-et-Marne consists of a mix of urban and rural areas and isrelatively large by French standards in terms of population (1.4millioninhabitants) as well as geographical area. [5]In France income-support programs are administered by governments of the départements on behalf ofthe federal French government. Given the institutional framework governingthe RSA regime, we are able to obtain and exploit data that are specific toSeine-et-Marne yet are nationally uniform in terms of reporting practices,organizational structure, and content.

11 The intervention itself consists of two fairly minor modifications of thecontent of the official letter which is sent to beneficiaries convoking them tomeet and confer with a case-worker. The measured outcome is the event ofparticipating in these counseling services. Generally we do not discern fromour findings a substantial increase in the participation rate of the treatmentgroups following the intervention. We do find, however, that in the cases oftwo particular groups, namely young males and individuals residing in ruralareas, our relatively minor intervention is associated with an increase in theparticipation rate by about10percentage points relative to that observed fora control group.

12 The second section of the paper provides some background on the institutional apparatus for the French social assistance regime and the phenomenon of low take-up. The third section presents the experimental design andthe empirical framework, and the final section presents the results and offersinterpretations.

30 The values presented in figure2represent the participation rates for theguidance interview cross-tabulated by experimental group along with confidence intervals at the90% level. Overall only55% of the households in thecontrol group participated in the guidance interview. Participation ratesappear to be higher for the two treatment groups, but the discrepancies arenot statistically significant.

31 The empirical framework consists of a linear probability regression modelin which the treatment status indicator is treated as randomly assigned. Theresults are presented in Table1as the differences in participation ratesbetween the control group and the two distinct treatment groups for theentire sample. [15]Control variables are added individually in the specifications listed in columns (1) through (4), and all of them are included in thespecification in column (5). [16]We note that the estimated coefficients of ourvariables of interest are very stable across the different specifications, andthat the estimated magnitude is higher for the second treatment. Comparedto the control group, the overall participation rates are higher for each of thetwo treatment groups, but these discrepancies are not statistically significant. This absence of significance at the conventional levels could be due toa lack of statistical power of the estimation. Given our relatively low samplesizes, we have less than a30% of chance of rejecting the null hypothesis atthe10% level for an estimated2percentage-point difference in the estimates between any two groups. Based on these findings, we assert that theimpacts of the treatments are not substantial for the entire sample.

32 We now turn our attention to an analysis based on less-aggregated controland treatment groups that are categorized by age, gender, and area of residence. These sets of results are presented in Tables2and3. The findings forthe first set of decompositions are shown in Table2. Note that since one ofthem is based on age, and we do not observe age for all subjects, the sizesfor those samples are somewhat smaller. We do discern a significant effectfor the second treatment group in the case of individuals residing in ruralareas: in this case the salient information mailing increases the probabilityof participating in the guidance interview by about10percentage points.This finding could reflect heterogeneous treatment effects that were maskedin the average treatment effects that were obtained for the entire sample.

33 Table3presents the results for the set of alternative decompositions. Theonly significant result that is discerned in this set of estimates pertains toyoung men in the case of the salient information treatment. This effectdisappears if the age threshold for inclusion in this subgroup is raised to30.Among this particular sample, the participation rate for that treatment groupis12.1percentage points (and22%) higher than the corresponding figurefor the control group. The participation rate of these households reachesalmost67% compared to55% for the control group. This finding could alsoreflect heterogeneous treatment effects that were masked in the averagetreatment effects that were obtained for the entire sample.

34 A possible interpretation is that this particular group is better positionedto search for employment due to factors such as health or family responsibilities (or lack thereof). Women tend to be more in charge of home production and familial responsibilities than men, which reduces their labour market activity (Châteauneuf-Malclès [2011]). This labour supply behavior tendsto increase with age and the number of children (DARES [2010]; Méda andPérivier [2007]). In contrast younger men might be more interested andinfluenced by this type of information than is the case for other inactive groups.The new information that is provided in the salient information letter mightlead some of them to re-evaluate upward the value of the option of participation in the guidance interview such that it offsets the costs of the multistep process about which some might be apprehensive. [17]

35 Linking the results to the framework presented in introduction, the experiment supports the relevance of a combination of lack of information andhigh perceived cost as one factor to explain the non-participation outcome.We note that it is not unusual to obtain results which vary by age and genderin RCTs involving the provision of varying information to subjects. Forinstance, Liebman and Luttmer [2015] conducted an RCT that dealt with theSocial Security Regime of the USA, which is the primary public retirementpension program in that country. When they varied the information that wasprovided to older (yet still active) workers regarding key provisions, theyuncovered a stronger impact among women than among men.

36 None of the other comparisons of the participation rates yield statisticallysignificant differences between the treatment groups and the control group.We discern no patterns for either the female samples or for the oldersamples (i.e., over28years of age). The simplified mailing is not associatedwith a statistically significant increase in the participation rate. We suggestthat failure to participate in the guidance interview does not appear to resultfrom misunderstanding the essence of any letter or inattention to the content of an overly-complex mailing. As explained above, the standard, original letter is already pretty brief, and it contains little more than a basicmessage—namely that the recipient only needs to contact their case-worker—which is tied very directly to our outcome variable. Recipients withserious cognitive biases, reading or comprehension difficulties may alreadyhave been flagged at the initial stage of their claim for RSA benefits and maytherefore be mostly absent from our sample of beneficiaries. For that reason, we exclude the possibility that recipients with those particular characteristics are reflected in our results.

37 As we noted above, due to our fairly small sample sizes and to the incidence of initial participation in the interview (about55%), which both generate a fairly high variance in estimation, the estimation procedures lackpower. The sub-samples were determined before the beginning of theexperiment, and thus our experimental design was constrained by the number of controls that were available. This in turn limits the increase in theprobability of committing type1error that is occasioned by multiple comparisons. The results are not significant when applying the very conservativeBonferonni correction for multiple comparisons. As a consequence, theresults based on sub-samples should be viewed as preliminary and exploratory in nature.

38 We can compare our results to those of other evaluative experimentswhich are also based on mailings. Our first intervention in particular iscommon to a number of studies. Our findings are in line with those generally found from experiments involving breast cancer screening, such asGoldzahl et al. [2018], who find no significant effect of their mailing interventions. The experiments conducted by Bourmaud et al. [2016] and Wardleet al. [2016] also failed to increase participation in screening by providingmore information about cancer in a supplemental leaflet. On the other hand,Bhargava and Manoli [2015] do find positive and significant effects (yet stillof only moderate magnitude) of their mailing experiment involving take-upof social assistance benefits in California.

39 This study consists of a field experiment that was conducted in order toevaluate an intervention that could increase the surprisingly low take-up ratefor social assistance benefits observed in France. We focused our attentionon a dimension of this topic that has received little attention in the literature,namely the disqualification of current recipients resulting from a failure tocomply with a critical regulation. About half of the income support beneficiaries whom we observe do not participate in the first step required in orderto maintain eligibility for receipt of the income support allowance, thusrunning the risk of benefit disentitlement. Our experiment consists of generating two informational “nudges” taking the form of minor and subtlevariations in the official letter that were sent to randomly selected groups.All of these letters notified beneficiaries of the obligation to present themselves at an interview with a case worker. The outcome variable for our RCTis the event of compliance.

40 Our findings indicate that the mailing that informed recipients about theadvantages (in the form of active labour market policy benefits) accompanying the passive income support benefits substantially increases the participation of two groups in the interviewing process, namely young men andindividuals living in rural areas. This result appears to be restricted to theseparticular samples, however, because the estimated effect is not significantfor either the entire sample nor for other subsamples. The other version ofthe letter, which consisted of a simplified version, does not appear to havestatistically significant effects on the variable of participation in the guidanceinterview. The apparently higher impact of the second treatment suggeststhat the lack of information as opposed to a surfeit of it plays at least apartial role in determining the outcome of participation.

41 Our findings suggest that the take-up of social assistance benefits,through the channel of avoiding disentitlement to benefits and thus ensuring continuous receipt, might be slightly increased by the simple and costless action of providing just a bit more information to recipients in regardsto extant but often unknown accompanying benefits. Depending on thenature of these highlighted benefits, the action may have a greater impacton certain segments of the population of beneficiaries. Our results suggestthat governments have little to lose by providing further information aboutthe set of benefits—typically taking the form of activation measures—accompanying the social assistance regime.

42 We note that in contrast to some of the existing research on the issues ofraising take-up and reaching out to and engaging beneficiaries, which isfocused solely on the behavior of beneficiaries, a policy repercussion of ouranalysis places some onus on the design and the administration of socialassistance programs. The quality and the content of the information transmitted to recipients and the response times of the administration likely playsome role in the unduly low take-up and unduly high disqualification phenomena. Ideally the information conveyed should be suited to the particularneeds and characteristics of the recipients. Indeed the objective of the guidance interview is to learn something about the recipient’s profile. In ourexperimental analysis, a small improvement or enrichment in the information that is initially provided to beneficiaries through formal letters mightfacilitate the implementation of a personalized “action” plan with an eyetowards re-integration into the labour market.

43 First on the agenda for further research is to obtain larger estimatingsamples that would reinforce the power of our statistical analysis. It wouldalso be highly desirable to conduct a survey of beneficiaries that measurestheir attributes, which could in turn be linked statistically to administrativedata on their characteristics and labour market outcomes. Such a processcould improve the targeting of both passive and active interventions.

44 As far as the external validity of our analysis is concerned, it applies tosocial assistance programs that offer related benefits and subject their beneficiaries to a meeting requirement. We do note that we cannot rule outother explanations for the high rate of non-compliance for the requiredinterview. An important but unknown factor is the degree to which the threatof suspension of benefit payment in case of non-participation is perceived ascredible by beneficiaries, as we do know that enforcement is far from guaranteed. It is possible that unobservable social network influences are at play.Another relevant explanatory variable of which we cannot take account isthe lags that occur between the time when the beneficiary does contact thecase-worker (which is not reported) and the time that the guidance interviewdoes take place. Subsequent research should attempt to address theseissues.