Tuesday, November 2, 2021

Shorter height and long-bone growth in the arms and legs were more evident among sao praphet song—who are both sexually oriented towards men and markedly feminine

An anthropometric study of sexual orientation and gender identity in Thailand. Malvina N. Skorska, Lindsay A. Coome, Diana E. Peragine, Madison Aitken & Doug P. VanderLaan. Scientific Reports volume 11, Article number: 18432. Sep 16 2021. https://www.nature.com/articles/s41598-021-97845-9

Abstract: The biodevelopment of psychological sex differentiation is putatively reflected in several anthropometrics. We examined eight anthropometrics in 1404 Thai participants varying in sex, sexual orientation, and gender identity/expression: heterosexual men and women, gay men, lesbian women, bisexual women, sao praphet song (transgender birth-assigned males), toms (transgender birth-assigned females), and dees (birth-assigned females attracted to toms). Exploratory factor analyses indicated the biomarkers should be analyzed independently. Using regressions, in birth-assigned males, less male-typical second-to-fourth digit ratios in the left hand were associated with sexual orientation towards men regardless of gender identity/expression, whereas shorter height and long-bone growth in the arms and legs were more evident among sao praphet song—who are both sexually oriented towards men and markedly feminine. In birth-assigned females, there were no clear sexual orientation effects, but there were possible gender-related effects. Groups of individuals who tend to be more masculine (i.e., toms, lesbians) showed more male-typical patterns on weight and leg length than some groups of individuals who tend to be less masculine (i.e., heterosexual women, dees). Thus, it appears the various anthropometrics inform separate biodevelopmental processes that differentially relate to sexual orientation and gender identity/expression depending on the measure in question as well as birth-assigned sex.


This study examined a comprehensive set of putative anthropometric biomarkers of brain and behavioral sex differentiation in the largest and most diverse sample to date. Using EFA, the several biomarkers previously independently associated with sexual orientation and/or gender identity/expression were reduced to three factors: body size (i.e., height, leg length, arm length, and weight had the highest loadings), hand ratio (i.e., right and left hand width-to-length ratios had the highest loadings), and digit ratio (i.e., left and right 2D:4D had the highest loadings). However, we were not able to demonstrate that these factors were invariant across groups, indicating that the manner in which these biomarkers relate to one another varies in relation to sex, sexual orientation, and gender identity/expression. Further, contrary to the possibility that the sex difference in 2D:4D reflects allometry38,39, lower 2D:4D among individuals assigned male at birth than those assigned female at birth was not associated with greater physical size with respect to the average length of the second and fourth digits, hand length, or height. Based on these observations, one cannot conclude that the various biomarkers examined here reflect some latent biodevelopmental process(es) influencing sex differentiation. Instead, the present analysis suggests they may each provide unique insights. Thus, examining individual biomarkers should be considered as a tenable approach when investigating their associations with traits such as sex, sexual orientation, and gender identity/expression.

Importantly, heterosexual sex differences were found for each of the individual biomarkers. Consistent with prior research in the West12,30,34,49, compared with heterosexual women, heterosexual men were taller, heavier, had longer arms and legs, had wider hands, and lower 2D:4D than heterosexual women. Thus, we were able to confirm that these biomarkers were sex-differentiated as expected among Thais, suggesting they may be useful to study in relation to sexual orientation and gender identity/expression in this population. Within-sex differences, where found, were generally consistent with the notion that developmental processes underlying the biomarkers are associated with sexual orientation and/or gender identity/expression—although the patterns of group differences varied by biomarker and by birth-assigned sex.

Among individuals assigned male at birth, height, the long bones, weight, and left 2D:4D were associated with group differences. Specifically, heterosexual men were taller and had longer arms and legs than sao praphet song. These differences could reflect differential androgenic effects on long bone growth at the prenatal and/or pubertal window(s) of development. Long bone growth is influenced in part by androgens acting on androgen receptors, and epiphysial closure is influenced by estrogens50,51. Sex differences in these biomarkers generally appear during puberty, with surges in androgens influencing their development during prenatal and pubertal development17. Social factors (e.g., stress, nutrition, social roles) have also been related to the development of these biomarkers17,18,19,36. Processes such as these that are tied to height and the growth of long bones within the prenatal and pubertal windows may, therefore, be implicated in these group differences. Gay men were intermediate relative to heterosexual men and sao praphet song on these measures but did not differ significantly from either group. This pattern could reflect a “dosage” effect, but such an interpretation is tentative in the absence of significant group differences. In any case, it appears that differences from heterosexual men in height and long bone growth are more evident among the androphilic birth-assigned males who are more markedly feminine in their gender expression (i.e., sao praphet song) in the current sample. In this respect, our findings parallel those of prior Western research suggesting smaller body size among androphilic birth-assigned males who are more feminine25,41. Our findings, however, do not align with other Western research that has found that gay men were shorter than heterosexual men20,21,22,24, although degree of femininity was not assessed in these samples.

Regarding weight, gay men weighed less than heterosexual men, supporting some previous studies in the West24,30. Furthermore, sao praphet song weighed marginally less than heterosexual men (p = 0.051, see Table S16), which aligns with the shorter stature of sao praphet song relative to heterosexual men, and providing some support for one Western study of transgender same-sex attracted birth-assigned males25. Gay men did not show significant skeletal differences (i.e., height, long bone growth) but nevertheless weighed less than heterosexual men, suggesting that this group difference likely resulted from differences in muscle and fat mass. Indeed, compared with heterosexual men, gay men are more likely to use diet pills, diet, purge, fast to lose weight, be dissatisfied with their appearance, and experience eating disorders52,53,54. However, the extent to which such tendencies also apply to gay men in Thailand is not known, and so this interpretation should be considered speculative.

We also found that, compared with heterosexual men, left 2D:4D was significantly greater among both gay men and sao praphet song, who did not differ significantly from one another. Digit ratio is argued to develop mostly under the influence of prenatal androgen exposure, given fetal 2D:4D sex differences55, but also via some genetic influence56, with no evidence indicating sociocultural influences10. The pattern of group differences suggests these biodevelopmental processes are linked to a sexual orientation effect whereby androphilia in birth-assigned males is associated with more female-typical digit ratio, regardless of whether gender expression is relatively feminine. As such, this pattern runs contrary to a recent study that reported digit ratio was more female-typical among gay men who expressed feminine gender role behavior57. Further, meta-analyses have suggested digit ratio is more female-typical (vs. heterosexual men) among trans women16, but not gay men12 (but see findings with an adolescent sample58). Given the discrepant findings, it will be important for future research to continue to examine digit ratio in relation to both sexual orientation and gender identity/expression.

Of note, the present study found group differences for left, but not right, 2D:4D. Prior research similarly found sex assigned at birth and sexual orientation digit ratio differences are more apparent in one hand than the other (i.e., either non-existent or smaller in effect size in one hand); however, effects have typically been more apparent on the right, not left, hand2,12,58,59. An exception is a Japanese study of digit ratio that reported a male sexual orientation difference on the left, but not right, hand60. Thus, although we did not find associations between digit ratio and male sexual orientation in both hands, such effects are commonly found in only one hand, or are found to be stronger in one hand, and the group differences observed in left, but not right, 2D:4D among individuals assigned male at birth in the current study are consistent with research in another Asian population. Reasons why this might be the case for certain populations requires further research.

Regarding hand ratios, we did not find any group differences among heterosexual men, gay men, and sao praphet song. As such, our findings did not replicate those of an earlier study that reported lower hand width-to-length ratios among gay, compared with heterosexual, men30. Of the anthropometrics that have been studied in relation to male sexual orientation, hand ratios have been examined seldomly, and to our knowledge have not been examined in relation to gender identity/expression. Further research is needed to determine whether hand ratios are likely to be informative of biodevelopmental processes influencing male sexual orientation and/or gender identity/expression.

Among individuals assigned female at birth, group differences on the various biomarkers did not correspond to differences in androphilic vs. gynephilic sexual orientation but instead tended to correspond to gender-related differences. Toms, who are more masculine-presenting than the other birth-assigned female participants, were heavier than heterosexual women, lesbians, and dees, who are all more feminine-presenting. There is some research suggesting that more masculine (butch) lesbians have greater circulating testosterone levels, higher waist-to-hip ratios, more masculine digit ratios, and greater recalled childhood gender-nonconforming behavior than more feminine (femme) lesbians and heterosexual women2,61,62. Thus, the weight result may support some role of androgens in the development of tom identity, although there was no support for a dosage effect and interpretative caution is warranted given the only difference in 2D:4D is opposite to what would be expected (see below).

We also found that toms and lesbians had longer legs than dees. Despite these group differences in leg length, there were no differences among the birth-assigned female groups in height, corroborating most previous findings suggesting no relationship between height and sexual orientation in females20,21,22, cf.23 and suggesting leg length may be the more relevant proxy to consider among females (also see30). The leg length pattern observed here might reflect that more male-typical leg length has a biodevelopmental association with attraction to feminine partners (as displayed by toms and lesbians) vs. masculine partners (as displayed by dees). That said, if such were the case, one would expect heterosexual women to show shorter legs as well given they are, relatively speaking, attracted to masculine men. Alternatively, these leg length differences may be related to gender role expression. In Thailand, the gender role behavior of toms and lesbians appears to be relatively more masculine than that of heterosexual women and dees, and dees are less masculine than heterosexual women48. Thus, group differences in degree of masculine gender role expression might account for why only dees and not heterosexual women had shorter leg length than toms and lesbians.

There were also some unexpected group differences among individuals assigned female at birth. First, contrary to the prediction that more masculine groups would show lower 2D:4D, toms had higher left 2D:4D than lesbian women. That said, lack of support for our prediction is not necessarily out of step with other literature given recent meta-analytic findings suggesting no differences in 2D:4D between heterosexual women and trans men16. As such, processes contributing to digit ratio might not be related to the development of masculine identity among individuals assigned female at birth. Second, heterosexual women had lower (more feminine) right-hand width-to-length ratios than dees. Hand development is thought to be influenced by androgens modulating specific homeobox genes63, with some evidence also pointing to hand use during childhood26,64. Given dees were the only group of female gynephiles who had more masculine right-hand ratios than heterosexual women, this finding provided relatively weak evidence of female sexual orientation being influenced by such mechanisms.

Overall, the current pattern of results for individuals assigned female at birth may support some role of elevated androgens in the development of toms and lesbians—although there was no clear support for a dosage effect and interpretative caution is warranted given several null differences from female comparators (e.g., lack of difference with heterosexual women in leg length). In any case, the body of evidence for a biological basis to the development of sexual orientation and gender identity/expression in females cannot be discounted2,65,66 and the present findings suggest that gender-related factors should continue to be assessed in biomarker studies of sexual orientation and gender identity/expression in individuals assigned female at birth. Moreover, further research examining cross-cultural (in)consistencies in biomarker expression patterns among birth-assigned females is needed. Given previous suggestions that sexual orientation and gender identity/expression are more fluid and/or influenced by sociocultural factors among birth-assigned females than males45,67,68,69, one might expect more inconsistency in biomarker patterns across populations among the former. In other words, there are potentially more factors beyond biological mechanisms of sex differentiation contributing to female, compared with male, sexuality and gender identity/expression. If so, among female groups within particular populations, these alternative factors may to some extent obscure group differences related to biological mechanisms.


Biomarkers provide an indirect assessment of the mechanisms purported to influence sex differentiation of the brain and behavior, including sexual orientation and gender identity/expression. Future research examining how these biomarkers relate to brain sex differences or, where possible, longitudinal studies that measure these mechanisms directly and link them to later behavioral outcomes would be valuable. Previous studies have shown measurement of 2D:4D and sex differences in 2D:4D to be impacted by indirect (e.g., photocopies) versus direct measurement70,71. Given we employed a direct method of measurement, group differences may be impacted in future replications of this work. Also, an EFA approach to studying a comprehensive set of biomarkers in both sexes and in relation to both sexual orientation and gender diversity has not been reported in studies of Western samples, making it somewhat difficult to compare the current EFA-based results to previous studies conducted with Western samples. Thus, replication of this approach in a Western sample is an important future direction.

Convenience and non-random sampling, primarily in an urban center, was utilized in the current study, which limits generalizability of our findings to the general Thai population, rural Thailand, or other non-Western cultures. We note, however, that although representative samples would be worthwhile to collect, these tend to suffer from small sample sizes of sexually and gender diverse participants21. Also, although the final sample size was comparatively large for studies in this literature, group sizes were relatively smaller for bisexual women and lesbian women, which might reflect that it is more normative in Thai culture for same-sex attracted females to identify with the categories of dees or toms rather than the more Western-style categories of bisexual and lesbian45. Other groups could not be included due to their small sample size (i.e., bisexual men, transgender men). We were unable to examine biomarkers in sao praphet song primarily attracted to women, or toms primarily attracted to men. These gaps may be due to cultural norms surrounding the gender identification and sexual preferences of third/nonbinary gender individuals within Thai society45. More targeted approaches to recruiting may facilitate broader and larger samples in which such groups are represented and would benefit the aim of disentangling sexual orientation from gender identity.

News exposure, twelve months worth of web browsing data: Exposure to partisan and centrist news websites—no matter if it is congenial or crosscutting—does not enhance polarization

No Polarization From Partisan News: Over-Time Evidence From Trace Data. Magdalena Wojcieszak et al. The International Journal of Press/Politics, November 1, 2021. https://doi.org/10.1177/19401612211047194

Abstract: Many blame partisan news media for polarization in America. This paper examines the effects of liberal, conservative, and centrist news on affective and attitude polarization. To this end, we rely on two studies that combine two-wave panel surveys (N1 = 303, N2 = 904) with twelve months worth of web browsing data submitted by the same participants comprising roughly thirty-eight million visits. We identify news exposure using an extensive list of news domains and develop a machine learning classifier to identify exposure to political news within these domains. The results offer a robust pattern of null findings. Exposure to partisan and centrist news websites—no matter if it is congenial or crosscutting—does not enhance polarization. These null effects also emerge among strong and weak partisans as well as Democrats and Republicans alike. We argue that these null results accurately portray the reality of limited effects of news in the “real world.” Politics and partisan news account for a small fraction of citizens’ online activities, less than 2 percent in our trace data, and are nearly unnoticeable in the overall information and communication ecology of most individuals.

Keywords: polarization, partisan media, online behavioral data, news exposure, media effects, affective polarization, attitude extremity, computational social science

Eviction policies: "Right-to-Counsel" drives up rents so much that homelessness increases by 15%, welfare is dampened, & default premia increase a lot; rental assistance lowers renters' default risk, reduces homelessness by 45% & evictions by 75%, & increases welfare

The welfare effects of eviction policies. Boaz Abramson. Standford Academic Market Program, Nov 2021. https://stanford.edu/~boaza/evictions_abramson.pdf

Abstract: This paper studies the implications of rental market policies that address evictions and homelessness. Policies that make it harder to evict delinquent tenants, for example by providing tax-funded legal counsel in eviction cases ("Right-to-Counsel") or by instating eviction moratoria, imply eviction and homelessness are less likely given default. But higher default costs to landlords lead to higher equilibrium rents and lower housing supply. I quantify these tradeoffs in a model of rental markets in a city, matched to micro data on rents and evictions as wel"Right-to-Counsel" drives up rents so much that homelessness increases by 15% and welfare is dampened. Since defaults on rent are driven by persistent income shocks, stronger protections are ineffective in preventing evictions of delinquent tenants, and lead to a large increase in default premia. In contrast, rental assistance lowers renters' default risk and as a result reduces homelessness by 45% and evictions by 75%, and increases welfare. Eviction moratoria can prevent a spike in evictions following a rare economic downturn, as long as they are used as a temporary measure.

1 Introduction

Approximately 2.2 million eviction cases are filed against renters every year (Desmond et al., 2018). A growing body of research documenting the negative outcomes associated with housing insecurity has triggered a public debate over policies that address evictions, as well as homelessness more generally. Policymakers across the country have considered enacting stronger protections against evictions, for example by providing free legal counsel in eviction cases (“Right-to-Counsel”), or by instating eviction moratoria. Rental assistance programs are also often proposed as a tool to promote housing affordability. While these policies provide a form of insurance to tenants who cannot pay rent, they can also affect equilibrium rents and housing supply. In this paper, I study the equilibrium effects of these policies. To this end, I propose a quantitative model of heterogeneous households who rent houses from investors, but can default on rent and face the risk of eviction. On the one hand, by making it harder to evict delinquent tenants, stronger protections provide greater insurance against idiosyncratic risk. On the other hand, they lead to higher equilibrium rents and lower housing supply because they weaken households’ ability to commit to paying future rent. I quantify the model using moments on evictions and rent, and then use it to evaluate the main rental market reforms that have been proposed. I find that “Right-to-Counsel” reduces evictions but drives up rents so much that it increases homelessness and reduces welfare. In contrast, means-tested rental assistance lowers tenants’ default risk, reduces both homelessness and evictions and increases welfare. The key force driving these differences are the dynamics of risk that underlie defaults on rent. When the shocks that lead to default are persistent, lawyers’ ability to prevent evictions and homelessness is limited because delinquent tenants are likely to continue defaulting until they are eventually evicted. At the same time, in a persistent risk environment, making it harder to evict delinquent tenants leads to relatively large increases in default premia and rents. Rental assistance promotes housing affordability and is welfare improving because it lowers default risk, as opposed to making it harder to evict tenants who have already defaulted. Consistent with this logic, I find that a temporary eviction moratorium following an unexpected unemployment shock, of the magnitude experienced at the onset of the COVID-19 pandemic, has little effects on rents and prevents a spike in evictions and homelessness. I identify the particular dynamics of risk that underlie defaults on rent using novel micro data on evictions. Using survey data, I document that the main risk factors driving defaults are job loss and divorce events, and that these shocks are associated with 1 persistent income consequences. By linking the universe of eviction cases to a registry of individual address histories that records demographic characteristics, I identify the populations that are at a particularly high risk to default and face eviction: young, less educated, and single households. I then provide evidence suggesting that these populations also face remarkably high and persistent income risk. The model accounts for these facts by explicitly modeling unemployment and divorces as sources of income risk, and by allowing the parameters of the income process to depend on age, human capital, marital status, as well as on divorce events. At the heart of the model are overlapping generations of households who have preferences over numeraire consumption and housing services and face idiosyncratic income and divorce risk. Households rent houses from investors by signing long-term leases that are non-contingent on future states. A lease specifies a per-period rent which is fixed for as long as the lease is ongoing. To move into the house, a household must pay rent in the same period in which the lease begins. The key feature of the model is that in subsequent periods households may stop making rent payments. When a household begins a default spell, an eviction case is filed against it. The eviction case extends until the household gets evicted or until it stops defaulting. Each period in which the household defaults it is evicted with an exogenous probability, which captures the strength of tenant protections in the city. A household who defaults but is not evicted lives in the house for free for the duration of the period, and accrues rental debt into the next period. Households with outstanding debt from previous periods can either repay the debt they owe, in addition to the per-period rent, or continue to default and face a new draw of the eviction realization. Guided by recent evidence on the consequences of eviction (e.g. Humphries et al., 2019), I model the cost of eviction as consisting of three components: temporary homelessness, partial repayment of outstanding debt, and a penalty on remaining wealth that captures, among others, health deterioration and material hardship that follow eviction. Houses are inelastically supplied by landowners to investors, who rent them to households. From the investors perspective, rental leases are risky assets. Investors incur a per-period cost for maintaining the house, but might not collect rents if their tenant defaults. Rental rates can depend on household observables and reflect the costs of default on rent to investors, such that in equilibrium investors break even. Houses are indivisible and there is a minimal size of housing. Households that cannot afford to move into the smallest house are homeless. The presence of a minimal house size reflects minimal habitability requirements and is consistent with the negative relationship between expenditure shares on rent and household income that I document from micro data. 2 In this setting, stronger tenant protections introduce more contingency in rental leases. They allow delinquent households to remain in their house for longer periods of time, thereby providing them with a better chance to avoid eviction and homelessness by repaying their debt later on. However, this increases the costs of default to investors, which leads to higher equilibrium rents and lower housing supply. Quantitatively, this trade-off depends on local rental market conditions. When renters’ income dynamics are such that the shocks that lead to default are transitory in nature, stronger protections can prevent evictions by providing delinquent tenants with more time to recover from a bad shock. However, if persistent shocks are the primary driver of defaults, protections are less effective because the shocks cannot easily be smoothed across time. The elasticity of housing supply in the city is also a key parameter for evaluating protections against evictions: when supply is less elastic, the effect on house prices is amplified. I quantify the model to the San Diego-Carlsbad-San-Marcos MSA, where homelessness is a major problem and high-quality eviction data are available. I specify an income process that allows for the distribution of both transitory and persistent components to depend on the household’s age, human capital, marital status, as well as on divorce events, and I estimate it to match the empirical evidence suggesting that households who are more likely to face evictions also face higher and more persistent income risk, and that divorces are both a driver of eviction and are associated with high income risk. I exploit detailed eviction court data to identify the strength of tenant protections in San Diego: the likelihood of eviction given default is identified by the average length of the eviction process, and the garnishment parameter governing debt repayment upon eviction is identified from the share of debt collected by landlords. I jointly estimate parameters with no direct evidence using a Simulated Method of Moments (SMM) approach. The estimation successfully matches facts on homelessness, evictions, rents and house prices. In particular, I estimate the minimal house size such that the average rent in the bottom housing segment matches the average rent in the bottom quartile of rents in San Diego. I identify the (dis)utility from homelessness from the homelessness rate in San Diego. The wealth penalty associated with eviction is identified from the eviction filing rate, which is the share of renter households who face an eviction case during the year. I am able to separately identify the homelessness service flow and the eviction penalty because both evicted households and those who do not sign a rental lease suffer from homelessness, but eviction carries the additional penalty. When homelessness is worse, both homelessness and eviction filings drop, but the eviction penalty shifts the two moments in opposite directions. A larger penalty disincentivizes default, but makes homelessness more attractive because staying out of the rental market elimi3 nates the risk of eviction. As a check of the model, I evaluate its fit to non-targeted moments. First, I show that while the model is estimated to match the overall eviction filing rate in San Diego, it also accounts for how eviction risk varies across households. The model matches the disproportionately high eviction filing rates observed for young households as well as the general downward trend across ages. It also does well in matching the share of eviction filings that are related to divorces. This is due to income data regularities, in particular younger households are poorer and are more likely to experience negative income shocks, and divorce is associated with elevated income risk. Second, I find that the model is qualitatively and quantitatively consistent with the empirical relationship between expenditure shares on rent and household income. I use the model to infer that the vast majority of defaults on rent in San Diego are instigated by persistent income shocks. In particular, 68% of default spells begin with a negative persistent income shock, 30% are due to a combination of both a negative persistent shock and a negative transitory shock, and only 2% are driven by a transitory shock alone. In this highly persistent risk environment, shocks cannot easily be smoothed across time, and there is limited scope for preventing evictions by making it harder to evict delinquent tenants. I then consider three policy experiments. In the first, I study the effects of instating a “Right-to-Counsel” reform. To do so, I employ micro level estimates on how legal counsel strengthens tenants protections against evictions. The “Shriver Act”, an RCT conducted by the Judicial Council of California in San Diego, finds that lawyers prolong the eviction process by approximately two weeks and lower debt repayments by 15% (Judicial Council of California, 2017). These estimates identify the parameters of the counterfactual eviction regime associated with “Right-to-Counsel”, namely a lower likelihood of eviction given default as well as a lower garnishment parameter. To evaluate the equilibrium effects of a city-wide “Right-to-Counsel” reform, I compute the steady state under these stronger tenant protections. The main result is that “Right-to-Counsel” drives up rents so much that homelessness rises by 15% in equilibrium. Since defaults are mostly driven by persistent shocks, lawyers are unable to prevent evictions (and homelessness) of delinquent tenants: the share of eviction cases that are resolved with an eviction (rather than repayment of debt) is nearly one in the baseline economy, and is only slightly lower under “Right-to-Counsel”. At the same time, the increase in default premia drives low-income households into homelessness. The eviction filing rate falls by 14%, but this is because the most risky tenants are priced out of the rental market. This result highlights that the evaluation of 4 tenant protections should take into account not only the effect on evictions, but also on housing affordability and homelessness. The reform also has distributional effects through its effect on house prices. As default premia increase, middle-income households are forced to downsize from upper to lower quality housing segments. This shift in demand for rentals leads to an increase in the house price in the bottom segments and a decrease in the house price in upper segments. Since rents partly reflect the price of housing, this amplifies the increase in rents driven by higher default premia in bottom segments, but mitigates the effect in upper segments. In fact, for high income renters in the upper segments who pose little risk for investors, rents are lower following the reform. In terms of welfare, I find that “Right-to-Counsel” dampens aggregate welfare. Welfare losses are particularly large for households at the bottom of the income distribution, who experience the largest increases in default premia. In contrast, rich renters experience rent declines and are better off. The second policy I consider is a means-tested rental assistance program that subsidizes $400 of monthly rent to households with income and savings below a threshold of $1, 000. I find that the program reduces homelessness by 45% and the eviction filing rate by 75%. Rather than making it harder to evict delinquent tenants, rental assistance lowers renters’ default and eviction risk. It also promotes housing affordability for households who previously could not afford to move into a house, both because it subsidizes their rent but also by lowering their default risk and therefore the rents they face. I find that the aggregate welfare effect of rental assistance is positive. However, effects differ across the population. Poor households who are eligible for the provision, in particular the young and those with human capital, are better off. However, households who are poor enough to rent low quality housing but not poor enough to be eligible for the subsidy are worse off because they pay higher rents. This is because the price of housing in the lower quality end increases, consistent with the common argument that rental assistance fuels demand for housing. Finally, I evaluate the program’s cost against an estimate of the savings in terms of expenses on homelessness, and find that the policy is cost-effective: it results in annual net gains of 8.3 million dollars to the San Diego MSA. I then evaluate the effects of enacting a temporary moratorium on evictions in response to an unexpected increase in the unemployment rate. I compute the transition dynamics following an unemployment shock of the magnitude observed in the US at the onset of COVID-19, for two scenarios: with and without a 12-month moratorium. I find that a moratorium significantly reduces homelessness and evictions along the transition path. This is because a large number of households who default on rent during the moratorium are able to repay their debt before being evicted. Compared to the “Right-to5 Counsel” reform, the moratorium successfully prevents homelessness because it provides delinquent households with a substantially longer period of time to repay their debt, and has little effect on rents because it is temporary in nature.

Who’s miserable now? Identifying clusters of people with the lowest subjective wellbeing in the UK

Who’s miserable now? Identifying clusters of people with the lowest subjective wellbeing in the UK. Paul Dolan, Kate Laffan & Alina Velias. Social Choice and Welfare, Nov 1 2021. https://link.springer.com/article/10.1007/s00355-021-01365-4

Abstract: Policymakers are generally most concerned about improving the lives of the worst-off members of society. Identifying these people can be challenging. We take various measures of subjective wellbeing (SWB) as indicators of the how well people are doing in life and employ Latent Class Analysis to identify those with greatest propensity to be among the worst-off in a nationally representative sample of over 215,000 people in the United Kingdom. Our results have important implications for how best to analyse data on SWB and who to target when looking to improve the lives of those with the lowest SWB.


In this paper, we define misery using the four measures of SWB used by the ONS. We consider someone to be in the most miserable group in society if they report low wellbeing on all four measures. In this way, we partly circumvent the debate about which of the four questions best reflects SWB and address concerns surrounding fuzzy preferences and simply mistaken subjective reports. According to this definition, 1.1% of the total sample are miserable. We examine who is among the worst-off in society by using LCA to identify groups of people united by specific observable characteristics and highlighting those characteristics that differentiate groups more vulnerable to misery from those at lower-than-average risk of being miserable.

The LCA highlights two groups that are at higher-than-average risk of being miserable. By far the most vulnerable are those belonging to class 1. Of the miserable people included in our analysis, class 1 account for 77%. Members of this group tend to be aged 30 + , economically inactive, face disability and health problems, live in rented accommodation, have compulsory or lower levels of education and tend not to be in a partnership. Those in class 2 are also vulnerable to misery, making up 19% of the miserable people in our sample. People in this class share some but not all of the characteristics which define class 1. Members of class 2 also report some health issues and have a higher-than-average risk of disability. They also tend not to be in a partnership. Unlike class 1, this group tends to be employed, is younger, more educated and is just as likely to have a mortgage as to be renting.

Together the members of these two classes make up just over 15% of the sample but they account for 96% of the most miserable members of society. These people, therefore, answer the question of who is miserable now. Their shared characteristics are perhaps unsurprising given some of the existing SWB literature. Many of the same characteristics that matter on average appear to be linked to misery too. Health, marital status and job security, for example, are long-established factors associated with SWB (Dush and Amato 2005; Steptoe et al. 2015; Dawson et al. 2017). The current work builds on existing studies by highlighting the substantive risk of misery facing those who concurrently lack a number of these different protective factors. Health literature is known to use clustering approaches to identify high and low health risk groups by looking at a combination of self-assessed, lifestyle and socio-demographic characteristics and propose tailored interventions (see e.g. Dodd et al. 2010)—and SWB literature can benefit from identifying misery-risk groups too.

Much of the existing literature has examined the determinants of LS. An analysis of the most miserable 5% of the population on LS yields similar results, with classes 1 and 2 remaining the classes which are the most vulnerable to misery. The major difference in the response to our overarching question of who is miserable when we look across the two definitions of misery, therefore, is one of scale rather than composition: Many more people are miserable when we define misery as low life satisfaction, compared to reporting low SWB across all four measures, but class 1 and 2 still account for the vast majority of the miserable in both cases.

We are aware that our approach is not without its limitations. In terms of identifying who are the most miserable, we must rely on the APS survey questions on people’s life circumstances and we must rely on those surveyed in the sample. The APS includes a broad range of questions, but it does not cover all of the dimensions of wellbeing of potential interest, nor all of the determinants of SWB that have been identified in the literature. For example, the APS is lacking indicators on people’s evaluation of the meaning of their lives and how people spend their time, which existing work identifies as an important dimension and predictor of SWB respectively (Stone and Mackie 2013; Laffan 2018). Furthermore, those interested in SWB and misery must do more to get at populations who do not participate in population surveys, such as the homeless and those in institutions such as care homes and prisons, many of whom we might expect to be among the worst-off in society. For example, homeless people, which, depending on the definition, constitute about 0.5% (320,000) of the UK population (Shelter 2018) and we do not capture them in our analysis.Footnote8

In terms of establishing the factors associated with who is the worst-off, LCA helps us to identify groups of individuals at the highest risks of misery but like most data science tools it requires large volumes of complete observations. This means that once a person fails to answer one of the survey questions (e.g., housing tenure) their entire entry is dropped from the clustering analysis, which can be a problem for the cases where the non-response to certain questions is group-specific (Heffetz and Rabin 2013). This can be particularly challenging if the non-response behaviour is correlated with the variable of interest, i.e., if the miserable tend to avoid answering certain questions about themselves.

We also cannot make causal claims based on our analysis. Like other correlational SWB research, the associations we present are vulnerable to reverse causality and omitted variable bias. As a result, insights from the current work do not suggest how to address people’s misery but rather identify those groups of people that policymakers should pay particular attention to. In particular, our results emphasise the importance of considering how and why individual factors may interplay to make people more or less vulnerable to misery. For example, the misery of those in poor health whilst active in the workforce may be driven by daily concern about job security. In contrast, the misery of those individuals whose poor health prevents them from participating in the workforce may be, in part, caused by the resulting loneliness they experience. Optimal policies to address misery should be informed by evidence on the way combinations of factors influence people’s SWB.

Even if the combination of characteristics that the analysis identifies as being predictive of misery do represent causal impacts on wellbeing, some characteristics will be more susceptible to policy intervention than others: job security compared to marital and disability status, for example. Several of the shared characteristics in both groups with a higher-than-average percentage of miserable—including a relatively high risk of being in poor health and having a disability—suggest that members of this group may be inelastic suppliers of wellbeing and the potential for policy intervention to improve their wellbeing may be limited.

Notwithstanding these limitations, the current work makes significant contributions to our understanding of who’s miserable now. One of the most important yardsticks for judging a society is how well it treats its worst-off. By looking across the four ONS wellbeing questions, we classify just over 1% of the APS sample as being in the most miserable group. By identifying which clusters of people are most vulnerable, we hope to have provided researchers and policymakers with insights which can assist them in more accurately identifying who to target when trying to improve the lives of the worst-off.