Americans overestimate the intergenerational persistence in income ranks

Prior Research on Perceptions of Economic Mobility

Intergenerational mobility indicates the openness, fluidity, and fairness in a society. Promoting the equality of opportunity has become a pressing policy concern, particularly in the recent era of rising inequality. The empirical literature on economic mobility is accompanied by a related literature on perceptions about mobility. The public perceptions about mobility are construed as an ideological condition under which attitudes of and preference toward inequality and redistributive policies are formed. How people view the status of mobility affects their openness or resistance to policy solutions that potentially reduce inequality and promote mobility (9, 10, 13, 20⇓⇓–23).

One strand of research suggests that the general public holds optimistic views on economic mobility in America (9, 11, 12). For example, Americans overestimate the probability that children starting from the bottom can make it to the top of the income distribution and underestimate the probability of a bottom-quintile child remaining at the bottom in America, while their Europeans counterparts hold more pessimistic perceptions about mobility in their countries (9). Another line of work, however, suggests that Americans are actually concerned about economic mobility (15⇓⇓–18) and that Americans have more enduring structural perceptions about mobility than widely assumed (18, 19). Furthermore, Americans’ concerns about the lack of opportunity have played a role in shaping their concerns about economic inequality (13, 24). Growth in economic inequality increases skepticism regarding the opportunity structure in society that, in turn, may motivate support for equity-enhancing policies (18). A detailed literature summary is presented in SI Appendix, section 1.

While previous work has taken important steps to understand the public perceptions of mobility, most of those measures tend to focus on the perceived transition probabilities for children from specific subgroups, such as the likelihood to move from bottom to top or from top to bottom (9, 12). However, these findings are sensitive to whether the socioeconomic spectrum is divided into 5 (25) or 3 (15) strata, whether phrases like “stayed in the bottom” or “moved up” were used (26), and the empirical benchmark against which people’s perceptions are compared (17). Another drawback is that these measures typically involve the calculation of probability, which is prone to cognitive errors by lay people, especially compared with average or expectation-based measures (16, 27, 28). Most importantly, these previous measures tend to focus on mobility chances for 1 or a few specific income groups, such as the top or bottom quintile. They have not examined the differences in perceived mobility prospects across the full spectrum of parent income ranks.

A Rank–Rank-Based Survey Instrument

This study contributes to current scholarship by proposing a survey instrument for measuring the variation in mobility perceptions, which is based on the relationship between parent and child income ranks. We present each respondent with 3 questions of the same format. The question first describes a hypothetical person whose parents’ income ranks at the pth percentile of the income distribution. A picture is also presented to illustrate the income ranking (SI Appendix, section 2). We then ask the respondent to drag a slider on a marked bar to indicate where she thinks this child’s own family income will rank at age 40. Each respondent will answer 3 such questions in which the hypothetical person’s parent income ranks (i.e., p) are drawn randomly without replacement from a total of 9 deciles (i.e., p=10,20,…,80,90). SI Appendix, section 2 presents the details about the survey instrument.

This survey instrument has several advantages. First, the question presents a reasonably tractable task to the respondents, as it avoids asking lay people to evaluate complicated transition probabilities. Second, as each respondent is asked to consider mobility outcomes for only 3 hypothetical individuals, each from a different parent income rank, the survey imposes a relatively light cognitive burden. These 2 advantages were confirmed in our pilot study, where we recruited 50 respondents via Amazon Mechanical Turk and asked them to describe their logic of reasoning in answering these questions (SI Appendix, section 3 has details and additional checks).

Most importantly, data collected from this instrument can be used to estimate the perceived rank–rank relationship, which can then be compared against the actual rank–rank relationship to describe the direction and degree of mobility misperception. To do this, for a sample of size n, we pool the answers across individuals to form a total of 3n (3 deciles per person × n persons) data points. Then, we plot the perceived and actual rank–rank lines against each other on the same plot for comparison. For both lines, a steeper rank–rank slope indicates greater intergenerational persistence in income ranks (i.e., lower mobility).

Fig. 1 illustrates 4 example cases for this comparison. In Fig. 1, solid (dashed) lines represent the actual (perceived) rank–rank relationship. The area Sover denotes the amount of overestimated mobility prospect, and Sunder denotes underestimated mobility prospect. A steeper perceived slope (βP) relative to the slope in reality (βR) indicates pessimism about the equality of mobility prospects between children in rich and poor families and vice versa.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 1.

Illustration of rank–rank lines in perception (dashed) and in reality (solid).

Case A presents an example in which the perceived rank–rank line is parallel to the actual line (βP=βR) but is placed at a higher vertical position at every level of parent income (Sover>0 and Sunder=0). This indicates that the public overestimates economic outcomes for children from all family backgrounds and that there are no misperceptions about the disparities in mobility outcomes. Case B presents an example in which the slope in the perceived line is steeper than the slope in reality (βP>βR), indicating pessimism about the equality of opportunity. In this case, people overestimate economic prospects for children from the upper one-half of the parent income distribution to the same extent as they underestimate economic outcomes for children from the lower one-half of the parent income distribution (Sover=Sunder). Hence, across the entire parent income distribution, the positive and negative misperceptions of economic prospects cancel each other out and generate a 0 average (Sover−Sunder=0).

The next 2 cases illustrate scenarios in which the public has misperceptions about the equality of opportunity (βP≠βR) and meanwhile, over- or underestimates economic prospects on average (i.e., Sover≠Sunder). Case C illustrates a scenario of pessimism about equality of opportunity (βP>βR). Meanwhile, the public overestimates economic outcomes for those at the middle- and higher-income ranks to a greater extent than they underestimate the economic outcomes for those at lower-income ranks (Sover>Sunder), leading to an overestimation about children’s economic prospects on average. By the same logic, Case D illustrates an overestimation of the equality of opportunity (βP<βR) but an underestimation of average economic outcomes (Sover<Sunder).

Two important features of this survey instrument are worth noting. First, in principle, Sover and Sunder should be equal, because on average, the amount of increase in income rank should be offset by the same amount of decrease. However, in the survey instrument, we do not give zero-sum instructions, because that would impose too much cognitive burden. Hence, it is possible that the respondents will overestimate (Cases A and C) or underestimate (Case B) the population-average income ranks, resulting in biases toward upward or downward mobility. An important advantage of our proposed measure is that the comparison between βP and βR is not prone to such bias, because the perceived rank–rank slope depends on the inequality in perceived mobility prospects rather than the average level. Second, because the parent income rank needs to be given in the question, there may be an anchoring problem: respondents’ answers may gravitate toward the parent income rank shown to them in the survey question, resulting in a bias toward the diagonal line in the perceived rank–rank line. In a robustness check, we removed the data points around the diagonal line. Results suggest that the anchoring effect is unlikely to affect our main conclusions (SI Appendix, section 6). We recommend that this robustness check be included in future applications of this survey instrument.

The key takeaway from the 4 example cases illustrated in Fig. 1 is that the problem of measuring the public’s (mis-)perceptions of mobility prospects across the parent income distribution can be reduced to a simpler problem of obtaining 4 parameters: Sover, Sunder, βP, and βR. Here, the first 3 can be estimated from the rank–rank-based survey instrument that we propose; the last parameter, βR, is obtained from estimates reported in large-sample empirical studies.

Data Collection

We collected data from a probability-based representative sample of 3,077 adults in the United States. Participants were recruited during February 2018 via National Opinion Research Center’s AmeriSpeak panel maintained by the Time-Sharing Experiments in the Social Sciences. We randomly assigned the respondents to 3 groups that vary in the attributes of the hypothetical person: (i) a baseline group in which there is no specific description (n = 1,858), (ii) a “college graduate” treatment group in which we describe the person as having a college degree (n = 606), and (iii) a “hard-working” treatment group in which we describe the person as working very hard in life (n = 613). These descriptions were added right after the information on the person’s parent income rank is given. Our analyses are weighted by AmeriSpeak panel weights. Weighted and unweighted sample statistics are presented in SI Appendix, section 4.

For the benchmark of actual mobility, we rely on the rank–rank relationship reported by recent work using tax records (1). These estimates do not suffer from measurement errors that are prevalent in survey data (29). Moreover, unlike other mobility measures, such as the intergenerational elasticity, which tend to be nonlinear, the rank–rank relationship in the United States can be well approximated by a linear line, the slope of which provides a succinct way for quantifying income mobility.

Empirical Results

Baseline Group.

As no additional information of the hypothetical person is given, results from the baseline group reflect the perceptions of overall mobility. Fig. 2 presents the conditional densities of child’s perceived income rank by parent income rank. The 3 horizontal lines at each parent income rank represent the 75th, 50th, and 25th percentiles of the child’s perceived income rank. While the child’s perceived income rank varies at each parent income rank, the median and the upper and lower quartiles of the perceived income rank distribution all increase steadily with parent income rank, showing intergenerational persistence in income ranks.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 2.

Conditional density distributions of child’s perceived income rank.

How well does the perceived rank–rank relationship match the reality? Fig. 3 contrasts the perceived rank–rank relationship in our data and the actual rank–rank line. For illustration, we have also plotted 2 auxiliary lines (Fig. 3, light blue dashed lines) indicating perfect income persistence (Fig. 3, diagonal 45° line) and perfect income mobility (Fig. 3, horizontal line where child’s mean rank = 50). A steeper rank–rank slope indicates greater closeness to perfect persistence, and a flatter slope indicates greater closeness to perfect mobility.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 3.

Rank–rank relationship in perception and in reality.

The pattern shown in Fig. 3 is akin to Case C in Fig. 1. The 2 lines cross at around the 34th percentile of parent income rank. To the right of this crossing point—that is, for children from families in the middle or upper parts of the income distribution—the public overestimates children’s income ranks. To the left of this crossing point, perceived economic ranks are lower than reality, indicating an underestimation of the economic prospects for children from the lower end of the income spectrum. Observe that Sover>Sunder, which implies that the public overestimates children’s economic prospects across their parent income ranks. Meanwhile, the slope of the perceived rank–rank line is steeper than the actual rank–rank line (0.65 vs. 0.34), indicating that the public perceives greater inequality in income ranks between children from rich and poor families than there actually is in reality.

Table 1 quantifies these results. Sover−Sunder captures the net difference between the child’s perceived and actual income ranks across the parent income distribution. The degree of pessimism about the equality of mobility prospects can be calculated as the difference between perceived and actual rank–rank slopes, which equals 0.31 (=0.65 − 0.34). That is, for a 1-unit increase in the parent income rank, the child’s expected income rank rises by 0.34 units in reality, but an average American expects it to rise by 0.65 units. That is, Americans overestimate the intergenerational persistence in income ranks. SI Appendix, section 6 presents additional results on perceived immobility, upward mobility, and downward mobility by parent income.

View this table:

• View inline
• View popup

Table 1.

Estimated model parameters and derived quantities

Between-Group Variations.

We next examine whether the public’s perceptions of mobility vary by respondents’ characteristics. To examine the between-group variations, we estimate a multilevel model predicting the child’s perceived income rank for respondent i and item k (=1, 2, and 3). Level 1 model isYikchild=β0iP+β1iP⋅Yikparent+ϵik,[1]

The random intercept (β0iP) and slope (β1iP) then depend on the respondent’s attributes X in the level 2 model:β0iP=Xγ0+ν0iβ1iP=Xγ1+ν1i.[2]

We present the coefficients predicting the slope in Fig. 4 and the full coefficient estimates in SI Appendix, section 7. Here, the effects of covariates on slope are of particular interest. A positive (negative) coefficient indicates greater (lesser) perceived inequality in mobility chances compared with the reference group. Respondents who are age 18–29, are white, are living in a household with $30,000–$99,999 annual income, own a house or pay some rent, lean liberal, or have a bachelor’s degree are more pessimistic about the equality of mobility prospects across parent income ranks.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 4.

Effects of respondent’s characteristics on perceived rank–rank slope (point estimates with 90% CIs). BA, bachelor’s degree; HS, high school.

Fig. 5 demonstrates the perceived rank–rank relationship by respondents’ education, household income, and political ideology. Higher education and income are associated with greater pessimism about mobility, a finding also noted by previous studies (9, 12, 25). Among those without a high school degree, the perceived rank–rank line is only slightly steeper than the reality; respondents with a bachelor’s degree are most pessimistic about the equality of mobility outcomes, particularly for children from the lowest 30% of income distribution. Respondents in all income groups are pessimistic about the equality of mobility prospects, and the perceived rank–rank slope is flattest among those with less than $30,000 annual household income.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 5.

Perceived rank–rank relationship by respondent’s education, household income, and political ideology.

Liberals are more pessimistic about the equality of mobility prospects, a finding consistent with prior findings (9, 15, 25). As discussed earlier, our rank–rank measure has the strength of allowing us to differentiate between perceptions of mobility prospects across parent income ranks. For example, here, the results suggest that respondents with different political orientation do not differ much in the perceived mobility for children with high parent income, but their perceptions differ substantially for children from the lower part of the income distribution.

Our analysis and interpretation above have focused on whether between-group differences exist rather than why they exist. The complex mechanisms behind these between-group differences are beyond the scope of this paper, and future research is needed to better understand the patterns of these differences. However, these results also suggest that Americans’ pessimism about the equality of mobility prospects across parent income ranks is quite robust across subgroups of the population.

College Graduate Treatment and Hard-Working Treatment.

Comparing the perceived rank–rank lines in the 2 treatment groups with the baseline group will tell us the extent to which the public perceives a college degree or hard work as a factor for promoting intergenerational mobility. Fig. 6 demonstrates the results. The college degree treatment raises the intercept (P = 0.000) and flattens the slope (P = 0.057) for the rank–rank line. When the hypothetical person has a college degree, respondents now perceive slightly higher-income ranks for children across the parent income spectrum, especially for those in the middle and lower parts. Consistent with the findings from empirical work (30, 31), the perceived rank–rank slope decreases when the hypothetical person has a college degree, but the perceived slope change (0.65–0.56) is much smaller than the change in reality (0.34–0.15) reported by recent empirical work (30). Hence, the public underestimates not only the amount of mobility but also, the contribution of a college degree to mobility. The perceived rank–rank line for the hard-working treatment has a slightly higher intercept (P = 0.086) than the baseline, and there is very small difference in their slopes (P = 0.955).

• Download figure
• Open in new tab
• Download powerpoint

Fig. 6.

Perceived rank–rank relationship in baseline and treatment groups.

Some nonlinearity of the rank–rank lines can be observed here: the 3 rank–rank lines are very close to each other at the upper 40% of the distribution but diverge for the lower 60%. Hence, we estimated the perceived slopes restricting to the lower 60% of parent income ranks. The changes in slopes are slightly higher but still very small. The pattern holds when we break down the sample by education and income (SI Appendix, section 8). These results suggest that the public remains pessimistic about the equality of opportunity even when the hypothetical person is assigned attributes that should promote economic mobility. The small treatment effects could also be due to the weakness of the intervention. The explanations behind the public’s pessimism about the role of college in promoting education warrant future research.

Discussion and Conclusion

The question of how Americans think about intergenerational mobility draws growing attention from scholars and policy makers who seek to understand its consequences for public opinion, political action, and policy support. In examining mobility perceptions, previous work has mainly focused on the perceived transition probabilities for children from specific positions in the parent income distribution. This study proposes a rank–rank-based survey instrument for measuring public perceptions of the inequality in mobility prospects. The key strengths of this measure are 3-fold. First, this measure differentiates perceived economic outcomes for children across parent income ranks, which enables us to measure the perceptions about the inequality in mobility prospects. Second, the survey item presents the respondents with a relatively simple, tractable task and requires little cognitive burden. Third, the perceived rank–rank line derived from the data can be directly compared with an up-to-date empirical benchmark based on administrative tax data.

In accounting for the differential perceptions of economic outcomes for hypothetical children from different parent income ranks, we have now come to form a more comprehensive picture of mobility perceptions. We conclude that Americans are pessimistic about the equality of economic prospects among children born to rich and poor parents. This finding challenges the prior belief that Americans hold an optimistic view about the openness of the society and is instead consistent with the idea that the everyday experiences of highly unequal opportunity between children from poorer and wealthier families may have shaped how the general public thinks about the opportunity structure of the American society. The results also indicate that the pessimism about the equality of mobility prospects results from their overestimation of income ranks for children from families in the middle and higher ends of the income distribution and their underestimation of income ranks for children from low-income families.

Reading our findings along with recent work on the public beliefs about inequality and mobility (13, 18), we suggest that the once-prevailing belief (that Americans tolerate inequality, because they see great promise in economic mobility) may be misleading. With large-scale, population-representative data, we show that Americans are aware of, and in fact quite pessimistic about, the remarkably unequal distribution of economic outcomes by family background. These findings call for an extended scope of inquiries that bridge the empirical research on economic mobility with research on the social, political, and psychological processes that shape how the public thinks about it. We also highlight the need for policy and political solutions to seriously engage with Americans’ concerns about the equality of opportunity in the society.