People’s attitudes towards poverty: it’s complicated (Part II)

In our last post, we talked about poverty narratives, and how people have complex views on poverty in the U.S. Here’s some insights so far:

  • People generally recognize structural barriers as drivers of poverty.
  • Most people do not believe that people living in poverty have character flaws when compared to the rest of the population.
  • People also generally believe that meritocracy and the American Dream are alive, and that working hard will lift people out of poverty.
  • Finally, people believe that some anti-poverty government programs require barriers to usage because users cannot be trusted not to commit fraud.

In this post, we’ll focus on how these views and attitudes vary (if at all!) depending on one’s sociodemographics and party affiliation. Specifically, we will look at income, education, age, race /ethnicity, religion, and party identification.

The bottom line is:

  • Gender, race, or ethnicity are not consistent predictors of poverty narrative endorsement (with a few exceptions we’ll discuss).
  • Education, income, and age group are somewhat more predictive. Overall, college-educated respondents and low-income respondents are less likely to endorse negative poverty narratives.
  • Partisanship and the degree to which religion is important in our lives matters much more. Religious respondents and those identifying with the Republican Party are more likely to endorse poverty narratives.

A quick note on the analysis: our survey items are shown on a 5-point scale ranging from strongly disagree (scored as 1) to strongly agree (scored as 5; you can check some of the items we used here). This post is based on the same data as our previous blog post. The plots you’ll see show the effect of each of our covariates on the level of endorsement for that item, given in standard deviations (e.g. in the first plot, Republican respondents score around 0.3 standard deviations higher in the items shown than Democratic respondents).

Personal narratives

Let’s start with the personal narrative items. Recall that people endorsing personal narratives believe that people living in poverty are more prone to personal failings, and that this is the ultimate cause of poverty.

We found that identifying as male and Republican were predictors of higher levels of endorsement for each of these items. Males and Republican respondents scored between 0.25-0.35 standard deviations (on average) higher in each of the statements in the personal narrative. The results for each of the predictors included are shown in the plot below.

Interestingly, gender is only a meaningful predictor for the personal narrative. Across all the other narratives analyzed, males and females barely differ.

Education, race, and income did not make a huge difference for these particular items (though we will later see that this depends on the narrative in question). The one exception is that Latino respondents were more likely to endorse the first item of the three items (“Poor people are dirty”), though this was not the case for the remaining two. Finally, note that older respondents scored lower than middle-aged and younger respondents across all items in the personal narrative.

Structural narratives

For structural narratives, which blame systems, institutions, or society as a whole for poverty, the largest predictors were party affiliation and income. Republicans scored over 0.5 SD lower than Democrats in the three structural items. Independents were also less supportive of this narrative; they were around 0.25 SD less supportive of the items shown below than Democrats in our sample.

The other big difference vis a vis the personal narratives was income. Those in the middle and top third of the income distribution* in our data were less likely to support structural items, whereas age was only a meaningful predictor for the last item. People aged 45 and above were less likely to endorse the idea that people living in poverty lacked opportunities for training.

Welfare Exploitation

As we mentioned last time, welfare exploitation narratives are a familiar trope, often used in the media to depict low-income people as people who take advantage of the safety net for their own benefit. This view holds that, at best, “welfare” (commonly thought of as public benefit programs) helps people but disincentivizes them from working hard; at worst, it attracts fraud and is a waste of money.

We saw the largest partisan differences in our welfare narrative items, perhaps because the debate around government programs and social assistance is highly politicized. There is also evidence that debates around “welfare” are racialized, with respondents being more likely to choose Black when asked to pick a typical “welfare recipient.”

The word “welfare” itself may convey partisan cues. Republicans scored over 0.75 SD higher than Democrats in their endorsement of the statement “welfare makes people lazy”. They were also much more likely to believe that there is a lot of fraud among government assistance recipients. While Independents were less likely to endorse these items than Republicans, they still scored around 0.3 SD higher than Democrats.

Respondents with a college degree (or more) were less likely to endorse welfare narrative items, though the effect sizes were smaller and noisier than for partisanship. Identifying as religious was also a positive predictor of endorsing welfare narrative items (and, conversely, identifying with no religion, agnosticism, or atheism predicted the opposite).


Finally, the patterns for the meritocracy narratives (the idea that poverty is a result of not working hard enough, and that most people can succeed if they put in the effort) are similar to those of the welfare exploitation narratives. Partisanship and religion are, once again, the key predictors, with income having a smaller effect.

Interestingly, age is predictive of our last meritocracy item. Older respondents are more likely to believe that the United States is, broadly, a meritocracy, and that everyone can attain the American dream.


The broad patterns are consistent: variables like race, ethnicity or gender are not (with some exceptions!) strong predictors of poverty narrative endorsement, whereas partisanship and the degree to which religion is important in our lives matters much more.

Education, income, and age group are also predictive, with college-educated respondents and those in the bottom third of the income distribution less likely to endorse negative poverty narratives. Interestingly, the influence of age-group varies more, with older respondents less supportive of individualist narratives, but more supportive of some welfare and meritocracy narrative items.

But what about other features? Could our worldview or other ideological constructs (such as authoritarianism, or social dominance) be predictive of our views on poverty?  The next blog post in our series will focus on precisely this; stay tuned for a piece on how ideology turns out to be the strongest predictor for overall attitudes towards people living in poverty.

Appendix: More Model Details

To generate the results above, a linear model (ordinary least squares) is used where the dependent variable is a score on a Likert scale (1 = strongly disagree to 5 = strongly agree) that has been standardized. For each model to explain what characteristics predict differences in support for a given item, controls are included for all the socio-demographic characteristics shown in the figures (age, income in terciles, education level, race, partisan affiliation, and reported importance of religion in your life). The reference categories for each variable are age group: 18-29, income: bottom third, education: high-school or less, self-reported race: white, party: Democratic, and religion: Not very important. Therefore, the effect of any individual characteristic can be interpreted as the impact of, for example, being in the top third of income, holding all other characteristics constant.  

*: Note that this refers to our sample. We divided our respondents’ household income data in thirds, with two thresholds at $40,000 and $70,000. This is similar, but not exactly the same as the United States income distribution.