Wednesday, May 12, 2021

More-men-more-violence association holds particularly for male violence against other men, but is insignificant for violence against women; significant among childless men, but not fathers; robustness checks question causality of associations

Are skewed sex ratios associated with violent crime? A longitudinal analysis using Swedish register data. Andreas Filser et al. Evolution and Human Behavior, Volume 42, Issue 3, May 2021, Pages 212-222.

David Schmitt's take: more-men-more-violence association holds particularly for male violence against other men, but is insignificant for violence against women...significant among childless men, but not fathers...robustness checks question causality of associations

Abstract: There is widespread concern in both the popular and academic literature that a surplus of men in a population intensifies mating competition between men, particularly unpartnered men, resulting in increased violence towards both men and women. Recent contributions challenge this perspective and argue that male mating competition and levels of violence will be higher when sex ratios are female-skewed. Existing empirical evidence remains inconclusive. We argue that this empirical ambiguity results from analyses of aggregate-level data, which put inferences at risk of ecological fallacies. Our analysis circumvents such problems by using individual-level, longitudinal demographic register and police data for the Stockholm metropolitan area, Sweden (1990–2003, n = 758,498). These data allow us to investigate the association between municipality-level sex ratios and violent offending (homicide, assault, threat, and sexual crimes) while adjusting for sociodemographic factors. Results suggest that aggregated offending rates are negatively associated with male-skewed sex ratios, whereas individual-level violent offending correlates positively with male-skews. We find that the more-men-more-violence association holds particularly for male violence against other men, but is insignificant for violence against women. Moreover, the association is significant among childless men, but not among fathers. However, robustness checks question the causality of these associations. Female violent offending is positively, albeit due to a low number of cases, insignificantly associated with male-skews. Moreover, both male and female non-violent offending is higher in male-skewed municipalities. We discuss the implications with regard to the theoretical debate and problems of unobserved heterogeneity in the sex ratio literature.

Keywords: Sex ratioViolent crimeMating marketSweden

4. Discussion

In this study, we use Swedish register data to investigate the link between local sex ratios and violent criminal offending. Existing theoretical approaches are contradictory and empirical evidence remains inconclusive, suggesting both negative and positive associations between male-skewed sex ratios and violence (Schacht et al., 2014Schacht et al., 2016Schnettler and Filser, 2015Schnettler and Filser, 2020). To our knowledge, this study is the first to use individual-level, longitudinal data to circumvent an ecological fallacy, a common problem of previous research on the issue (Filser & Schnettler, 2018Pollet et al., 2017Schacht et al., 2014). The detail of our data enable us to disentangle associations of intra- and inter-sexual violence with the sex ratio and to adjust for a number of socio-demographic confounders. Moreover, we investigate hypothesized, but largely untested differences in these associations by socio-economic, marital, and parental status.

On the surface, our results seem to provide additional support for the more-men-more-violence hypothesis. We find that male-on-male violence is positively associated with male-skewed sex ratios. Results for male-on-female violence suggest a similar positive association with male-skews, which is somewhat weaker and not statistically significant, due to a smaller sample size for these offenses. Both findings are compatible with the hypothesis that an abundance of men will particularly result in high levels male-male violence. Furthermore, our results suggest that intra-male violence correlates significantly with local sex ratios among childless men, but not among fathers. This may appear as further support for the hypothesis that local male-skews particularly instigate violent rivalry among men competing for partners.

However, the full set of results, including the results from models examining female-on-female violent offenses, and non-violent offenses, suggests caution before drawing any firm conclusions. First, when comparing our findings on male and female offending, we would expect either no association between sex ratios and female offending or one that is opposite to the association with male offending (cf. Stone, 2015). Yet, we find that the association between the sex ratio and female violent offending resembles that for male offending. The association for female offending is not significant, but this may be largely due to the low number of offenses by women in our data. The number of offenses is even lower for analyses of female offending split by victims' sex, which precludes any meaningful interference from results on female-on-male and female-on-female violent offending.

Second, we find that municipality-level sex ratios are not only associated with violent offending, but also with a general indicator of any non-violent offending. Property and white-collar offenses might correlate with sex ratios, as individuals are more pressurized to obtain resources and resort to scramble competition (Benenson & Abadzi, 2020Edlund et al., 2013). However, given the broadness of the indicator, we would expect these associations to be weaker, compared to violent offenses. While this is only true for non-violent offending by men, we find that the association of sex ratios with female non-violent offending is even stronger than the one for violent offending. Furthermore, we find that the associations of sex ratios with non-violent offending are in the same direction for both male and female offenders. This contradicts theoretical expectations related to scramble competition as women should become less and not more likely to engage in non-violent offending as the sex ratio increases, that is, as the environment becomes less female-skewed.

In sum, these findings prompt us to suspect that there are still potential confounders that might drive the association between sex ratios and violent crime that we are not able to account for. We are able to adjust our models for contextual and individual-level socio-economic deprivation in a more comprehensive way than previous studies. Socio-economic deprivation is a key confounder of the association between sex ratios and violent crime, given that young women are more likely to migrate to more economically thriving regions (Leibert, 2016) and levels of violent crime are correlated with prosperity (Hooghe et al., 2011). We address this issue by including municipality-level fixed effects to account for time-constant unobserved heterogeneity on the municipality level. Moreover, we include a range of time-varying socioeconomic and demographic status variables on both the individual and context level. These adjustments should take care of socio-economic unobserved heterogeneity, yet some limitations remain.

Beyond socioeconomic factors, sex-selective migration patterns might be a source of unobserved heterogeneity. Empirical evidence suggests that women out-migrate from male-biased areas to more strongly female-biased areas than men (Uggla & Mace, 2017). With regard to violence and crime, one potential explanation could be that men are less concerned about falling victim to a crime (Jackson, 2009). Consequently, sex-selective migration might drive the association of sex ratios and violent crime independently of economic deprivation. Unfortunately, we are not able adjust our models for migration patterns in our analysis, particularly migration from outside our study area.

Another limitation of our study is the comparatively small geographical scope of our data. An underlying assumption of our analysis is that individuals are sensitive to cues of the municipality-level sex ratio and that municipalities meaningfully represents the local ecology which impacts individual behavior. However, individuals might have committed offenses in different contexts than their municipality of residence, resulting in a mismatch of the contextual sex ratio at the offense and our sex ratio measure. Moreover, municipalities might be too small entities to measure sex ratios in a way that also correlates closely with individuals' perceptions of partner markets (Filser & Preetz, 2020Fossett & Kiecolt, 1991Gilbert, Uggla, & Mace, 2016). Furthermore, our study area consists of a metropolitan area with an urban center, Stockholm city. Municipalities are a meaningful social entity in Sweden, because they organize schools and municipality centers serve as local hubs. However, individuals still commute and move between municipalities. With a size of 7150 km2, the area is well connected by public transport and roads. Consequently, municipalities are not as separate as calculating specific municipality-level sex ratios suggests them to be.

Nevertheless, the level of detail in our data allow us to elucidate a number of aspects previous studies have not been able to explore. A key contribution of our paper is to support concerns about studying the association of sex ratios and aggregated rates of individual social outcomes, as it is commonly done in the existing sex ratio literature (cf. Pollet et al., 2017 for an in-depth critique). Specifically, our findings demonstrate how, based on the same data, sex ratios and aggregated violent offending rates can suggest a negative association, even when individual probabilities for violent offending are actually positively associated with sex ratios. This illustrates the importance of individual-level analyses to further establish a coherent empirical basis in the sex ratio literature.

Moreover, our paper illustrates that detailed offending data are necessary to generate clearer evidence with regard to which types of violent offenses are associated with sex ratio skews. We wish to remain cautious with too much emphasis on the differential levels of significance due to diverging sample sizes in offenses. Yet, the weaker association for male-on-female offending compared to male-on-male offending puts predictions of increased male-on-female intimate partner violence (D'Alessio & Stolzenberg, 2010Daly & Wilson, 1998Uggla & Mace, 2015bVandello, 2007) or higher levels of sexual harassment in male-skewed environments into perspective (Malamuth et al., 2005Trent & South, 2012Trent, South, & Bose, 2015). However, our data do not differentiate between violence against (intimate) partners and other female victims and thus our results on male-on-female violence can only serve as a combined indicator of violence against women by both partners and other men. Additional research is necessary to disentangle these different types of male-on-female violence and their associations with the sex ratio.

Moreover, our results reveal that the association of skewed sex ratios with violence may differ across individual demographic characteristics, as has been shown for other outcomes (Uggla & Mace, 2017). Specifically, sex ratios are positively associated with violent offending in childless men, but not among fathers. While we outlined above that this finding should be treated with caution, it still serves as an illustration for the yet untapped potential of individual-level data for the literature on sex ratios and violent offending. Future studies should further explore this aspect.

In sum, our findings demonstrate the need for studies relying on more detailed data and advanced causal identification strategies when exploring the association of sex ratios with violence and aggression. Observational studies, no matter how detailed, might be unable to overcome unobserved heterogeneity problems at both the individual and the aggregate level. Experimental studies have generated evidence for sex ratio effects on aggression by human participants (Arnocky, Ribout, Mirza, & Knack, 2014). We encourage future research to evaluate whether these effects vary across parental status groups as our results indicate.

Finally, our findings should be considered within their specific socio-sexual context (see Schacht et al., 2014). Our study population fits the WEIRD definition (Henrich, Heine, & Norenzayan, 2010), with high levels of acceptance for uncommitted sexual relationships (cf. Widmer, Treas, & Newcomb, 1998). Therefore, our paper complements the literature in that it comes from a sexually liberal society, while previous individual-level analyses use data from more sexually restrictive contexts (Diamond-Smith & Rudolph, 2018South et al., 2014). Such societies might not be suitable test cases for the more-men-more-violence hypothesis, since this perspective emphasizes uncommitted sexual relationships as a main mediator for the link between sex ratios and male violence (Schacht et al., 2014Schacht et al., 2016). This limitation does not apply to our study. While we cannot provide evidence of a counterfactual causal effect of male-skewed sex ratios on violent crime, our findings at least cast doubt on the more-men-less-violence hypothesis vis-à-vis the more-men-more-violence hypothesis in this context.

CRED Louvain climate disaster events “This is creating weather-related disasters that are completely unprecedented”

Dot Earth


Gore Pulls Slide of Disaster Trends
BY ANDREW C. REVKIN FEBRUARY 23, 2009 12:31 PM February 23, 2009 12:31 pm 183
COLELLAPHOTO.COM via Al Gore addresses the American Association for the Advancement of Science.

Former Vice President Al Gore is pulling a dramatic slide from his ever-evolving global warming presentation. When Mr. Gore addressed a packed, cheering hall at the annual meeting of the American Association for the Advancement of Science in Chicago earlier this month, his climate slide show contained a startling graph showing a ceiling-high spike in disasters in recent years. The data came from the Center for Research on the Epidemiology of Disasters (also called CRED) at the Catholic University of Louvain in Brussels.

The graph, which was added to his talk last year, came just after a sequence of images of people from Iowa to South Australia struggling with drought, wildfire, flooding and other weather-related calamities. Mr. Gore described the pattern as a manifestation of human-driven climate change. “This is creating weather-related disasters that are completely unprecedented,” he said. (The preceding link is to a video clip of that portion of the talk; go to 7th minute.)

Now Mr. Gore is dropping the graph, his office said today. Here’s why.

Two days after the talk, Mr. Gore was sharply criticized for using the data to make a point about global warming by Roger A. Pielke, Jr., a political scientist focused on disaster trends and climate policy at the University of Colorado. Mr. Pielke noted that the Center for Research on the Epidemiology of Disasters stressed in reports that a host of factors unrelated to climate caused the enormous rise in reported disasters (details below).

Dr. Pielke quoted the Belgian center: “Indeed, justifying the upward trend in hydro-meteorological disaster occurrence and impacts essentially through climate change would be misleading. Climate change is probably an actor in this increase but not the major one — even if its impact on the figures will likely become more evident in the future.”

Officials at the disaster center, after reviewing what Mr. Gore showed and said, sent a comment to Dr. Pielke’s blog and to me. You can read their full response below. I sent it to Mr. Gore’s office and asked for his interpretation. Kalee Kreider, Mr. Gore’s spokeswoman on environmental matters, wrote back today:

I can confirm that historically, we used Munich Re and Swiss Re data for the slide show. This can be confirmed using a hard copy of An Inconvenient Truth. (It is cited if you cannot recall from the film which is now several years old!). We became aware of the CRED database from its use by Charles Blow in the New York Times (May 31, 2008). So, it’s a very new addition.

We have found that Munich Re and other insurers and their science experts have made the attribution. I’m referring you particularly to their floods section/report [link, link] Both of these were published in a series entitled “Weather catastrophes and climate change-Is there still hope for us.”

We appreciate that you have pointed out the issues with the CRED database and will make the switch back to the data we used previously to ensure that there is no confusion either with regards to the data or attribution.

As to climate change and its impacts on storms and floods, the IPCC and NOAA among many other top scientific groups have indicated that climate change will result in more extreme weather events, including heat waves, wildfires, storms and floods. As the result of briefings from top scientists, Vice President Gore believes that we are beginning to see evidence of that now.
Reproduced with permission, from Guha-Sapir, D., Vos, F.; Quantifying Global Environmental Change Impacts: Methods, Criteria and Definitions for Compiling Data on Hydro-Meteorological Hazards in Coping with Global Environmental Change, Disasters and Security – Threats, Challenges, Vulnerabilities and Risks. Edited by H. Brauch et al Hexagon Series on Human and Environmental Security and Peace, vol. 5 (Berlin – Heidelberg – New York: Springer-Verlag, 2009). Click on the image for a full-sized version of the graph, which shows storm trends as a green band and flooding and related landslides in blue.

At right is a link to the Belgian center’s graph of disaster trends through 2008 (click to see the whole thing). And here is the center’s statement (highlight added):

CRED is fully aware of the potential for misleading interpretations of EM-DAT figures by various users. This is a risk all public datasets run…. Before interpreting the upward trend in the occurrence of weather-related disasters as “completely unprecedented” and “due to global warming”, one has to take into account the complexities of disaster occurrence, human vulnerabilities and statistical reporting and registering.

Over the last 30 years, the development of telecommunications, media and increased international cooperation has played a critical role in the number of disasters that are reported internationally. In addition, increases in humanitarian funds have encouraged reporting of more disasters, especially smaller events. Finally, disasters are the convergence of hazards with vulnerabilities. As such, an increase of physical, social, economic or environmental vulnerabilities can mean an increase in the occurrence of disasters.

We believe that the increase seen in the graph until about 1995 is explained partly by better reporting of disasters in general, partly due to active data collection efforts by CRED and partly due to real increases in certain types of disasters. We estimate that the data in the most recent decade present the least bias and reflect a real change in numbers. This is especially true for floods and cyclones. Whether this is due to climate change or not, we are unable to say.

Once again, we would like to point out that although climate change could affect the severity, frequency and spatial distribution of hydro-meteorological events, we need to be cautious when interpreting disaster data and take into account the inherent complexity of climate and weather related processes — and remain objective scientific observers.

[UPDATE: 5:10 p.m.: I’ve posted on a more measured effort at climate risk communication.]

Also on the disaster-climate front, there is an interesting story in the Washington Post today describing a variegated assemblage of efforts to flee in the face of climate-related threats. Matthew Nisbet pondered how a global warming story without a hot political element made it onto a front page.

It’s pretty clear it was the climate-disaster link. There were some things missing from the article, however. As the folks in Belgium explained above, the connection between human-driven climate change and recent trends in disasters remains highly uncertain, even as most climate scientists foresee intensification of floods and droughts and, of course, more coastal flooding with rising sea levels.

So while the climate hook might have given this story its “front-page thought,” there’s no examination in the article of simultaneous trends in population growth in poor places, urbanization (people are leaving marginal lands for many reasons) and the like.

In the absence of that hook, it’s basically a story about people moving out of harm’s way, something that’s been happening throughout human history.183

From 2019... Middle‐aged women with a greater number of recent stressful life events demonstrate memory decline over a decade later

Stressful life events and cognitive decline: Sex differences in the Baltimore Epidemiologic Catchment Area Follow‐Up Study. Cynthia A. Munro  Alexandra M. Wennberg  Nicholas Bienko  William W. Eaton  Constantine G. Lyketsos  Adam P. Spira. International Journal of Geriatric Psychiatry, March 22 2019.


Introduction: The reasons why women are at higher risk than men for developing dementia are unclear. Although studies implicate sex differences in the effect of stress on cognitive functioning, whether stressful life events are associated with subsequent cognitive decline has received scant research attention.

Methods: In Wave 3 (1993–1996) of the Baltimore Epidemiologic Catchment Area study, 337 men and 572 women (mean age = 47 years) reported recent (within the last year) and remote (from 1981 until 1 year ago) traumatic events (eg, combat) and stressful life events (eg, divorce/separation). At Waves 3 and 4 (2004–2005), they completed the Mini Mental State Examination (MMSE) and a word‐list memory test. Multivariable models were used to examine the association between traumatic and stressful life events at Wave 3 and cognitive change by Wave 4.

Results: A greater number of recent stressful life events at Wave 3, but not of more remote stressful events, was associated with greater verbal memory decline by Wave 4 in women but not in men. Stressful events were not associated with change in MMSE, and there were no associations between traumatic events occurring at any time and subsequent memory or MMSE decline in either sex.

Conclusions: Unlike men, middle‐aged women with a greater number of recent stressful life events demonstrate memory decline over a decade later. Sex differences in cognitive vulnerability to stressful life events may underlie women's increased risk of memory impairment in late life, suggesting that stress reduction interventions may help prevent cognitive decline in women.

From “NASA Lies” to “Reptilian Eyes”: Mapping Communication About 10 Conspiracy Theories

From “Nasa Lies” to “Reptilian Eyes”: Mapping Communication About 10 Conspiracy Theories, Their Communities, and Main Propagators on Twitter. Daniela Mahl, Jing Zeng, Mike S. Schäfer. Social Media + Society, May 12, 2021.

Abstract: In recent years, conspiracy theories have pervaded mainstream discourse. Social media, in particular, reinforce their visibility and propagation. However, most prior studies on the dissemination of conspiracy theories in digital environments have focused on individual cases or conspiracy theories as a generic phenomenon. Our research addresses this gap by comparing the 10 most prominent conspiracy theories on Twitter, the communities supporting them, and their main propagators. Drawing on a dataset of 106,807 tweets published over 6 weeks from 2018 to 2019, we combine large-scale network analysis and in-depth qualitative analysis of user profiles. Our findings illustrate which conspiracy theories are prevalent on Twitter, and how different conspiracy theories are separated or interconnected within communities. In addition, our study provides empirical support for previous assertions that extremist accounts are being “deplatformed” by leading social media companies. We also discuss how the implications of these findings elucidate the role of societal and political contexts in propagating conspiracy theories on social media.

Keywords: conspiracy theory, social media, Twitter

Conspiracy theories are a fast-changing phenomenon and highly responsive to external events. In light of the ongoing COVID-19 pandemic, a plethora of conspiracy theories abound online. Going beyond previous studies on either the general phenomenon of conspiracy theories (e.g., Del Vicario et al., 2016) or specific conspiracy theories (e.g., Broniatowski et al., 2018), our study provides an empirically informed comparison of the most visible conspiracy theories on Twitter by shedding light on the interplay of platform affordances and the dissemination of conspiracy theory content.

Regarding the diversity of conspiracy theories, our results reveal a variety of prevalent conspiratorial explanations circulating on Twitter: Agenda 21, Anti-Vaccination, Chemtrails, Climate Change Denial, Directed Energy Weapons, Flat Earth, Illuminati, Pizzagate, Reptilians, and 9/11. While most of these conspiracy theories are directed against the establishment and elite, referring to secret machinations of influential people or institutions acting for their own benefit (e.g., Agenda 21, Illuminati), others construct narratives challenging science, epistemic institutions, or scientists (e.g., Anti-Vaccination, Flat Earth).

Concerning communities evolving around conspiracy theories on Twitter as well as main propagators within these communities, our results reveal two loosely connected clusters of pro- and anti-conspiracy theories. Both anti-conspiracy theory communities, Anti-Flat Earth and Pro-Vaccination, are centered around scientists and medical practitioners. Their use of pro-conspiracy theory hashtags likely is an attempt to directly engage and confront users who disseminate conspiracy theories. Studies from social psychology have shown that cross-group communication can be an effective way to resolve misunderstandings, rumors, and misinformation (e.g., DiFonzo, 2013). By deliberately using pro-conspiracy hashtags, anti-conspiracy theory accounts inject their ideas into the conspiracists’ conversations. However, our study suggests that this visibility does not translate into cross-group communication, that is, retweeting each other’s messages. This, in turn, indicates that debunking efforts hardly traverse the two clusters.

Finally, our study lends support to previous assertions that social media platforms are taking increasingly proactive measures to systematically crack down on accounts promoting conspiracy theories (e.g., Rogers, 2020). As our results show, users banned from Twitter are predominantly those who propagate conspiracy theories.

Alignments Between Conspiracy Theories and Their Communities

In line with recent research demonstrating that conspiracy beliefs tend to “stick together” (Douglas et al., 2019, p. 7; van Prooijen, 2018), our study reveals a general proximity between several conspiracy theories. We argue that three factors help us to explore these overlaps in more depth. First, the closely aligned conspiracy theories Climate Change Denial, Pizzagate, and 9/11 share structural and thematic features. They provide alternative rationales and explanations for national or international policies, political affairs, or events, challenging accounts from governments and official authorities (Huneman & Vorms, 2018Räikkä, 2009). Other conspiracy theories such as Chemtrail, Reptilians, and Illuminati emphasize a conspiracy of powerful groups and non-human entities, claiming that, for instance, aliens or secret societies rule the world (Uscinski, 2018). In contrast to political conspiracy theories, these conspiracies mingle reality with fiction and are often closely tied to popular culture such as Dan Brown’s novel “The Da Vinci Code.” Second, factors such as ideological and geographic proximity further help explain alignments between conspiracy theories. A shared characteristic of several conspiracy theories is that they are disseminated by people with conservative political views who support Donald Trump and that they are mostly popular in the United States. For instance, conspiracy theories around Climate Change, Agenda 21, and Directed Energy Weapons have their roots in anti-environmentalist and anti-globalist ideologies, both of which are aligned with conservative political ideology and values (Kirilenko & Stepchenkova, 2014) and popular among the political right and populists in the United States (Harris et al., 2017). 9/11 and Pizzagate conspiracy theories are also widely promoted by conservative and right-wing politicians in the United States and often supported by individuals who hold conservative beliefs (Stempel et al., 2007). In contrast, believers in Anti-Vaccination, Flat Earth, Reptilians, Illuminati, and Chemtrail conspiracy theories can be found across the political spectrum, and their respective communities are less US-centric. For instance, Anti-Vaccination sentiment is on the rise around the globe and the movement finds supporters on both the political left and right (Holt, 2018). In addition, some of the most influential propagators of Illuminati and Reptilians conspiracy theories are based outside the United States (Robertson, 2013), which underlines the importance of societal and political contexts to understand the propagation patterns of conspiracy theories.

Limitations and Future Research

As all studies, ours comes with some limitations as well. A general limitation resides in the way we built our sample. First, hashtag-based approaches to collect tweets leave out ancillary discussions by participants who have chosen not to use these hashtags or any hashtags at all (Burgess & Bruns, 2015). Future studies should dive deeper and make use of alternative sampling methods, such as including specific user-defined keywords, utilizing topic-related dictionaries or classifiers, or examine recent tweeting history and follower network information of participating accounts to capture further communication (Burgess & Bruns, 2015). Second, our analysis was limited to a relatively small sample of English-language Twitter only, limiting the generalizability of our findings. As prior research suggests that conspiracy theories are communicated differently according to national and regional contexts (e.g., Gray, 2008), studies on other languages and linguistic regions would be recommendable.

To further enhance our understanding of conspiracy theories in digital environments, future research should incorporate more cross-platform, cross-lingual, and cross-regional comparative perspectives in general. Furthermore, we argue that future research of online conspiracy theories should not be limited to mainstream platforms, such as Twitter, Facebook, or YouTube. These platforms, as indicated in both literature (e.g., Rogers, 2020) and our current study, have been systematically cracking down on accounts that promote conspiracy theories. As more and more conspiracy theorists and their followers migrate to “alternative” social media, such as Gab, BitChute, and Parler, more research will be required to investigate the impacts of this trend. 

Since it is difficult to hold politicians accountable for personal welfare changes, to protect our self-image we tend to take personal responsibility for positive changes & hold the government responsible for negative changes

How Do Voters Hold Politicians Accountable for Personal Welfare? Evidence of a Self-Serving Bias. Martin Vinæs Larsen. The Journal of Politics, Volume 83, Number 2, May 2021.

Abstract: Examining a government’s record is difficult. This is a problem for voters who want to hold governments accountable. One solution is for voters to hold governments accountable for changes in their personal welfare. Yet, it is often unclear whether changes in personal welfare are caused by government policies or voters’ own actions. Since voters have a desire to protect their self-image, this ambiguity might fuel a self-serving bias in attribution. That is, voters might take personal responsibility for positive changes in personal welfare and hold the government responsible for negative changes. Using data from election surveys and survey experiments, this article shows that voters attribute responsibility for personal welfare in this self-serving way. This hurts democratic accountability because voters do not reward governments (enough) for improving their personal welfare.