Poverty Prefers Company

Overview of Studies

We tested our prediction in three web-based experiments that allowed us to examine both actual poverty and an experimental manipulation of (imagined) poverty. The first two studies employed the same two dependent measures. First, we had participants indicate their ideal percentage of poor in a society. Second, participants indicated whether they would prefer to join a nation with 34% poor people or with 10% poor people. Given a definition of “poor” as a family receiving an income of less than half of the national average, these income distributions correspond to the United States and Sweden, respectively.

To address whether preferences correlated with actual poverty, we asked participants to which income category they belonged. To establish the hypothesized causal relationship between poverty and preferences for inequality, we also introduced a between-subjects manipulation. In one condition, participants were asked to state their national income distribution preferences assuming that they would be poor. Control conditions in the first study asked participants to imagine either that they would be rich or that they would not know which income category they would belong to. In the second study, the control condition simply asked participants to imagine that they would belong to their current (actual) income category.

A third study was designed to disentangle the poverty prefers company effect from an alternative, everyone-prefers-company, effect. To this end, Study 3 respondents made choices between four pairs of national income distributions where the proportions or poor and rich people were manipulated independently.

Study 1

Method

An online survey was completed by representative samples of 1,058 American (51% female, mean age = 40) and 1,060 Swedish respondents (50% female, mean age = 42). ¹ Each respondent was assigned to one of the three conditions described below. All respondents completed six measures in the following order:

Preferred Income Distribution

We presented income distributions as the relative sizes of four categories, explained to respondents as follows:

Poor: individuals whose income is less than half the national average (i.e., income less than 50% of average income).

Lower middles: those with incomes between 50% and 100% of the national average.

Upper middles: those with incomes between 100% and 200% of the national average.

Rich: those with incomes greater than 200% of the national average.

Pie charts displayed the income distribution in two nations (Figure 1). Depending on condition, respondents were asked to imagine that, whichever of these nations they joined, they would either “be in the poor category” (imagine-poor condition), “be in the rich category” (imagine-rich condition), or “be randomly assigned to a place in the distribution” (veil-of-ignorance condition).

OPEN IN VIEWER

Figure 1.

The two income distributions presented in the survey, modeled on Sweden (left) and the United States (right).

Results

Preferred Income Distribution and Ideal Percentage of Poor

Recall that we measured preferences for inequality in two ways: the choice of preferred income distribution (United States/Sweden, dummy coded 1/0) and the ideal percentage of poor. As expected, the two measures were correlated (U.S. sample: r = .21, p < .001; Swedish sample: r = .23, p < .001).

Table 1 reports descriptive statistics per condition and country for these measures. The table shows two clear patterns: First, preferences for inequality were greater in the U.S. sample than in the Swedish sample. More relevant to our theoretical arguments, preferences for higher levels of inequality were also greater in the imagined poverty condition than in either of the two control conditions. ² Table 2 reports logistic and linear regression analyses for the binary and continuous preference measures, respectively, establishing the statistical significance of these patterns. ³ These results were unchanged when a range of control variables—age, gender, college education, and political views—were included. (These analyses are available upon request.)

OPEN IN VIEWER

Table 1.

Descriptive Statistics of Dependent Measures (Study 1).

Country	Condition	N	Preferred the U.S. Income Distribution (%)	M (SD) Ideal Percentage of Poor
United States	Imagine-poor	166	35.5a	21.2a (24.7)
	Imagine-rich	168	24.4b	14.7b (19.5)
	Veil-of-ignorance	173	17.3b	13.4b (16.7)
	Total	507	25.6	16.4 (20.8)
Sweden	Imagine-poor	185	22.2a	10.9a (18.0)
	Imagine-rich	179	11.7b	6.9b (11.9)
	Veil-of-ignorance	176	6.8b	5.7b (9.0)
	Total	540	13.7	7.9 (13.7)

Note. Entries in a given column with different subscripts significantly differ according to Holm–Bonferroni post hoc tests.

OPEN IN VIEWER

Table 2.

Logistic Regression Predicting Preference for the U.S. Income Distribution and Linear Regression Predicting Ideal Percentage of Poor (Study 1).

Factor	Preference for United States	Ideal Poor Percentage
United States	2.27***	0.24***
Imagine-poor	3.05***	0.17***
Imagine-rich	1.62*	0.03

Note. Entries are odds ratios from logistic regression (left column; Nagelkerke R ² = .08) and standardized coefficients from linear regression (right column; adjusted R ² = .08).

*p < .05. ***p < .001.

A binary measure of actual poverty was obtained from participant’s income category by coding poor participants as 1 and others as 0. We also analyzed the effect of actual poverty in the veil-of-ignorance condition where participants were not told to which income category they would belong. In the U.S. sample (N = 173, of which 42 were poor), 33% of poor respondents preferred the more unequal income distribution compared to just 12% of participants in the nonpoor categories, an odds ratio of 3.59, p = .002. In the Swedish sample (N = 176, of which only 20 were poor), 10% of poor respondents preferred the more unequal income distribution compared to 6% of participants in the nonpoor categories, an odds ratio of 1.62; however, the extremely small cell size makes statistical tests meaningless. Unexpectedly, actual poverty did not correlate with ideal percentage of poor in either country, a point to which we return in the discussion below.

Discussion

In our first study, only a small minority (12%) of American respondents who were not poor preferred the more unequal U.S.-type income distribution to the Swedish one. However, among poor respondents, there was a substantially larger tendency to prefer the more unequal distribution. We found similar effects for the experimental manipulation: Compared to those in the veil of ignorance condition, American respondents who were asked to imagine that they would be poor were more than twice as likely to express a preference for the highly unequal U.S. distribution. This tendency was even larger among Swedish respondents, who expressed a 3-fold increase in their preferences for the more unequal system when asked to imagine that they would be poor. These results strongly support our proposition that poverty prefers company.

The experimental manipulation yielded similar effects for respondents’ ideal percentage of poor. For this measure, however, we did not see a correlation with actual poverty under a veil of ignorance. It is likely that participants’ current income level was insufficiently salient in that condition, since they were asked to imagine that they would be randomly assigned to a place in the distribution. A second limitation of the first study was the large exclusion rate, due to respondents’ limited comprehension of wealth presented as percentiles (see Eriksson & Simpson, 2012). Our second study sought to replicate the Study 1 findings with a design that circumvented these two concerns.

Study 2

Method

An online survey was completed by a sample of 550 American adults (47% female, mean age = 32; 56% with college education). Participants, recruited from the Amazon Mechanical Turk (mturk.com), were assigned to one of two conditions described below.

Our first goal was to make income categories easier for respondents to understand. Thus, in Study 2, we defined income categories in terms of absolute levels, as recent work shows substantially higher comprehension rates using this approach over the percentile approach of Study 1 (Eriksson & Simpson, 2012). That is, Study 2 defined household income in absolute terms as follows: poor: less than $25,000; lower middles: $25,000–$50,000; upper middles: $50,000–$100,000; rich: more than $100,000. Respondents were shown the same pie charts as in Study 1 (Figure 1). They were then asked to indicate their own household’s income category, before completing the same measures of preferred income distribution and ideal percentage of poor as in Study 1. When completing these measures, respondents were asked to imagine that they would either “be in the poor category” (imagine-poor, N = 200) or “be in the same category you stated you belong to today” (imagine-actual, N = 350). ⁴

Finally, to assess understanding of income distributions given in the pie diagrams, respondents were asked to indicate which of the two nations has “the least unequal income distribution (i.e., fewer poor people and fewer rich people)?”

Results

Preferred Income Distribution and Ideal Percentage of Poor

As in Study 1, preference for the more unequal income distribution was correlated with higher ideal percentage of poor, r = .46, p < .001. Table 3 reports descriptive statistics for these two measures of preferences for inequality. Compared to the imagine-actual-income condition, the imagine-poor condition had a higher proportion of participants preferring the more unequal income distribution (odds ratio 3.06, p < .001) and a higher average ideal percentage of poor (r = .16, p < .001). Within the imagine-actual-income condition (N = 312), actual poverty was similarly related to a higher proportion of participants preferring the more unequal income distribution (odds ratio 2.75, p = .008) and a higher average ideal percentage of poor (r = .15, p = .008). All these results were essentially unchanged when control variables were included (analyses available upon request).

OPEN IN VIEWER

Table 3.

Descriptive Statistics of Dependent Measures Per Condition (Study 2).

Condition	N	Preferred the More Unequal Income Distribution (%)	M (SD) Ideal Percentage of Poor
Imagine-poor	172	26.1a	13.9a (15.9)
Imagine-actual	312	12.9b	9.5b (10.3)

Note. Entries with different subscripts in a column significantly differ.

Discussion

Study 2 provided additional support for the poverty prefers company effect. However, the first two studies do not allow us to distinguish a tendency for poor people to prefer company from a more general effect of everyone preferring company. Specifically, just as imagining membership in a poor category might create a preference for a larger poor population, it might be that imagining membership in the rich category also creates a preference for a larger rich population. An additional weakness of the first two studies was potential ambiguity about the consequences of category size. Participants might have assumed that a larger poor population larger would be better off than a smaller one. For instance, they might have assumed, for whatever reason, that each individual poor person would be better off when there were many others than when there were fewer others (e.g., because a larger share of societal resources would be devoted to helping the poor). Or they might have assumed that a large population of poor would be more apt to benefit from collective action or other changes aimed at improving the lot of the poor. We designed the third study to address these issues.

Study 3

Method

A total of 300 American adults (37% female, mean age = 31; 69% with college education), recruited from the Amazon Mechanical Turk, participated in Study 3. Participants were randomly assigned to one of the three conditions: imagine-poor, imagine-rich, or imagine-actual-income.

The definition of income categories and comprehension checks of pie charts followed those of Study 2. Respondents were then presented with four new pairs of income distributions, presented in pie charts. Each distribution had the sizes of the rich and poor categories at either a low level (9%) or a higher level (15%). The middle categories were sized such that the median was always the cut point between the lower and upper middles. Specifically, the four pairs were as follows:

Pair 1: (9% poor: 41% lower middle: 41% upper middle: 9% rich) versus (9%: 41%: 35%: 15%).

Pairs 1 and 2 test preferences for proportions of rich when the proportion of poor is fixed at a low (Pair 1) or high (Pair 2) level. Pairs 3 and 4 test preferences for proportions of poor when the proportion of rich is fixed at a low (Pair 3) or high (Pair 4) level:

Pair 3: (9%: 41%: 41%: 9%) versus (15%: 35%: 41%: 9%).

For each pair, respondents indicated which nation they would rather join, or if they were perfectly neutral. To remove ambiguity about possible consequences of category size, the instructions stated: “When indicating your preference, assume that the standard of living for any given category member will remain constant, regardless of the size (or number of members) of that category. Further assume that there will no policies or redistribution efforts that would impact the standard of living of category members.” For all choices, respondents were asked to imagine that they would either “be in the poor category” (imagine-poor), “be in the rich category” (imagine-rich) or “be in the same category you currently belong to” (imagine-actual-income). Participants remained in the same income condition (poor, rich, or actual) for all four choices.

Results

Preferred Income Distribution

Each of the four choices was conceived as a choice between a high or low level of the manipulated category (either rich or poor). Responses were scored 1 for preference for high level, 0.5 for neutral, and 0 for preference for low level. The two measures for preference for many poor were then averaged to a single preference-for-many-poor index (Cronbach’s α = .58), and similarly for the two measures for preference for many rich (α = .49). Preferences for many rich and many poor were uncorrelated (r = −.03, p = .57). Descriptive statistics for the two preference indices for each of the three conditions are reported in Table 4.

OPEN IN VIEWER

Table 4.

Descriptive Statistics of Dependent Measures (Study 3).

Condition	N	M (SD) Preference for Many Poor	M (SD) Preference for Many Rich
Imagine-poor	94	.25a (.36)	.36a (.37)
Imagine-rich	95	.09b (.22)	.38a (.39)
Imagine-actual-income	97	.14b (.29)	.44a (.38)

Note. Entries in a given column with different subscripts significantly differ according to Holm–Bonferroni post hoc tests.

An analysis of variance (ANOVA) showed that preferences for many poor differed significantly between the three conditions, F(2, 283) = 6.88, p = .001. Consistent with Studies 1 and 2, as shown in Table 4, the preference for many poor was higher in the imagine-poor condition than in any of the other conditions.

In contrast, preferences for many rich did not differ significantly between conditions, F(2, 283) = 1.07, p = .35. Specifically, the preference for many rich was no different in the imagine-rich condition than in the other conditions. ⁵ The statistical significance of the interaction between condition (imagine rich, poor, or actual income) and measure (preference for rich vs. poor) was confirmed in a three-by-two mixed ANOVA, F(2, 283) = 4.46, p = .01.

Discussion

Study 3 replicated results from Studies 1 and 2 using a design that allowed us to address two important alternative explanations: that the findings were driven by an “everyone prefers company” effect and that poverty prefers company because larger categories of poor promise better conditions for the poor. The findings from Study 3 strongly support the poverty prefers company prediction, allowing us to rule out these alternatives.

General Discussion

In this article, we have explored whether poverty might prefer company, a possibility suggested by prior work on resource distributions in small groups (Eriksson & Simpson, 2011; Kuziemko et al., 2011). Three web-based studies supported this hypothesis. When participants imagined being poor, their preferences tended to shift toward greater inequality. Similarly, when participants imagined they would stay in their current income category, poor participants tended to prefer greater inequality than participants in higher income categories.

On the surface, our findings might appear to run counter to Piff, , Kraus, Côté, Cheng, and Keltner’s (2010) recent demonstration that the poor tend to be more egalitarian and, as a consequence, more generous than their counterparts from higher social classes. But the ostraphobia (Eriksson & Simpson, 2011) and last place aversion (Kuziemko et al., 2011) arguments on which we based our predictions posit an aversion to being alone at the bottom of a rank or distribution. Thus, while Piff and colleagues posit a linear (and negative) relationship between social class and helping, our findings deal specifically with members of the bottom social class. Moreover, even among those at the bottom, a preference that others are similar in status need not imply less willingness to engage in the routine, everyday helping behaviors of the type investigated by Piff et al. More generally, offering to care for a peer’s child or providing a temporary monetary loan to those in similar circumstances does not generally, if ever, improve the beneficiary’s social class position. In short, our arguments and findings complement, rather than contradict recent findings in the social psychology of social class.

Importantly, our first study showed that the poverty prefers company effect held among citizens of both the United States and Sweden. Among Western nations, the United States and Sweden have very high and very low levels of inequality, respectively. We found that Swedes and Americans tend to have quite different notions of “ideal” levels of inequality, suggesting that individuals’ preferences tend to adapt to the realities of the society in which they live. Yet, manipulating (imagined) poverty had similar effects on Americans and Swedes’ preferences for inequality. Moreover, these effects were independent of respondents’ political views, age, education, or gender. This robustness across and within cultures suggests that the poverty prefers company effect may be relatively universal.

While the goal of our current work was to demonstrate the poverty prefers company effect, rather than an explicit mechanism, Study 3 allowed us to narrow down the possible causes of this effect. To begin with, the effect remained even when instructions ruled out any implicit assumptions that larger classes of poor will be more successful at reducing inequality or otherwise improving their lot. Further, results of orthogonal manipulations of the sizes of the poor and rich categories confirmed that the effect is not driven by a tendency for everyone to prefer company. Rather our findings suggest that an increase in the preferred size of one’s own income category is specific to the poor.

Why might the poor be especially apt to prefer company? One possibility is that being part of a larger category allows for external versus internal attributions about one’s circumstances. For instance, when many are poor, I can tell myself (and others) that society offers too few people opportunities to get ahead. When fewer are poor, I may be forced to question whether I have the talents and abilities required for success. Another, noncompeting, possibility is suggested by recent work on last place aversion (Kuziemko et al., 2011) or ostraphobia (Eriksson & Simpson, 2011) which shows that members of small groups who receive the smallest rewards prefer redistributions where other group members receive equally small rewards, even though such redistributions tend to enhance overall inequality in the group.

Whatever the exact cause, we believe that the poverty prefers company effect has important implications. For instance, contrary to the intuition that poor people would be the strongest supporters of policies to reduce poverty, our findings suggest that poverty may make people less, not more, interested in reducing the number of poor people. If so, efforts to mobilize the poor to reduce inequality might be more effective when they are framed as efforts to improve the lots of individuals rather than reducing the number of poor as a whole.