Relatively accurate but absolutely off: U.S. residents’ estimates of relative and absolute economic mobility

ABSTRACT

Using results from two treatment arms of a larger nationwide survey experiment in the United States, we add evidence on how Americans perceive social mobility. In one arm, respondents estimate both upward and downward relative economic mobility, while in the other, respondents estimate the absolute economic mobility (that is, do children out-earn their parents?) for each quintile of the parental income distribution. This latter question has been overlooked in the literature. We find respondentsʻ average estimates of relative downward mobility to be remarkably close to reality. The average estimate of relative upward mobility is also close, though both estimates underestimate relative mobility. There are small partisan differences in accuracy on relative mobility, with Republican-leaning respondents providing the most accurate estimates, but respondentsʻ perception of absolute mobility significantly underestimate (overestimate) mobility for the poorest (highest-income) children irrespective of partisan affiliation. Our results provide important insights on the implementation of redistribution policies.

KEYWORDS:

• Economic mobility
• social mobility
• survey
• lay beliefs

JEL CLASSIFICATION:

• C93
• D63
• D72
• H23
• J62
• J68

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1. Though see Alesina, Stantcheva, and Teso (2018) and Barton and Pan (2021) for evidence contrary to this hypothesis.

2. Often researchers use a more stringent cutoff than any improvement in relative rank. Below, we follow Davidai and Gilovich (2015a) and define relative upward (downward) mobility as children born to the bottom (top) quintile who move up (down) more than one quintile as adults.

3. Swan et al. (2017) and Davidai and Gilovich (2018) demonstrate that some, but not all, of the differences in results are due to methodological differences.

4. An anonymous reviewer offers an alternative perspective: that contra Davidai and Gilovich (2016), predicting relative mobility is the easier of the two estimates, as it only requires some knowledge of the social dynamism within an economy, while predicting absolute mobility requires both this knowledge and understanding of long-term per capita economic growth. While we have not explored this idea formally, its prediction is clearly consistent with our findings.

5. Our other paper examines the role of information provision and argumentation on support for six social-mobility-linked public policies. We find information on relative and absolute mobility do not affect support for any policies, but several argument treatments significantly increase support for their associated policies. We also allowed respondents the opportunity to write their U.S. Senators, and find argument treatments significantly increase the likelihood that letters address economic mobility and significantly promote advocacy for the associated policy in the letter, but we find no increase in advocacy from the information treatments. Please see Barton and Pan (2021) for details.

6. Flynn, Nyhan, and Reifler (2017) argue that misperceptions are driven by motivated reasoning. In contrast, Landy, Guay, and Marghetis (2018), in a discussion of misperceptions about demographic compositions of different nations, argue that the best explanation for such systematic misestimation has less to do with bias or availability (e.g., in media coverage or representation) and more to do with human cognitive processes common to all questions of estimating under uncertainty.

7. We compare our question on “family” income to the average U.S. household income, as it is not clear that most respondents live in families as defined by the U.S. Census Bureau. Our average income in the sample ($66,000) is between the 2018 U.S. median household income ($61,937) and U.S. median family income ($76,401).

8. We present results here with unweighted data. At an anonymous reviewerʻs suggestion, we examined reweighting the data. We reweighted by sex, education, and age, and separately by sex, race, and age. Reweighting makes our sample more representative along the characteristics chosen, but less representative by income in both cases. Our main result—that Americans hold relatively accurate views of relative mobility when compared to their views on absolute mobility—remains unchanged. Results are available upon request

9. The survey is available upon request.

10. Like Alesina, Stantcheva, and Teso (2018), Chambers, Swan, and Heesacker (2015), and Swan et al. (2017) (but unlike Cheng and Wen 2019), we indicate the income quintile by family income at birth, but do not explicitly indicate whether the childrenʻs end point (relatively or absolutely) is in terms of family or individual income. And like all these papers, we do not explicitly raise the issue of pre- vs. post-tax income. While it is possible that this lack of specificity has impacted respondentsʻ answers, we doubt this for two reasons. First, as the title of the paper indicates, respondentsʻ views of relative mobility are in fact quite close to the actual data from Chetty et al. (2014). And second, when it comes to absolute mobility in study 2, Chetty et al. (2017) demonstrates that the patterns of absolute mobility are very similar whether one uses family or individual income. Thus, while we think it would be interesting to examine how increasing the precision of the question may improve (or reduce) the accuracy of respondentsʻ beliefs, we leave that exercise to a future paper.

11. An anonymous reviewer asks us to discuss how this question, which only allows for linear combinations of “luck” and “effort”, addresses other characteristics that may affect life outcomes (e.g., background, discrimination, or ability). We did not provide any guidance to respondents on the question, but we expect that many use “luck” for factors outside an individualʻs control, and “effort” for factors within an individualʻs control. That said, as even personality characteristics, such as conscientiousness, that affect “effort” are heritable (and thus outside a personʻs control) (Luciano et al. 2006), we can only say for sure that those who view success as “effort-based” hold different views on mobility than those who view it as “luck-based”.

12. Respondents answer on a 0-10 scale. We pool the five most luck-oriented categories in Figure 4, as only 10 percent of respondents chose an answer of 4 or below, but use the full scale in our regression. Our regression results are insensitive to using the truncated scale.

13. Because the average response across quintiles here is nearly identical, one may ask whether this pattern might be due to hasty or inattentive subjects selecting the same answer for each quintile. We recreated the figure with only those respondents in the top 75 percent of the distribution of completions times (those who took 4.15 minutes or longer to complete the survey); the pattern remains unchanged. We also calculated the within-subject standard deviation of estimates across the five quintiles. The average standard deviation was 16.8 (median 16.4), and only two subjects had no variation in responses. The flat distribution of absolute mobility estimates appears driven primarily by the wide distribution of respondentsʻ beliefs. We thank the reviewer for the question.

14. One may ask whether respondents estimating the middle quintile correctly indicates that they understand absolute mobility on. To address this, we average respondentsʻ answers across quintiles, and compare this estimate of average mobility to average mobility from Chetty et al. (2017). While numerically close, respondentsʻ average (44.3 percent) statistically significantly underestimates actual mobility (47.3 percent; p = 0.005, two-tailed t-test). We thank the reviewer for raising this issue.

15. It also does not appear that respondents thought it was true for absolute mobility and attempted to make their answers sum to 100: the median of the sum of absolute mobility estimates across quintiles is 215, and only 21 of 266 respondents have absolute mobility estimates that sum to less than 100 percent.

16. The test statistics (J*) for Republicans, Democrats, and independents (respectively) for the bottom 20 percent are −11.752, −17.574, and −13.567. For the top 20 percent, they are 9.232, 14.823, and 10.33. In all cases, we can reject both no ordering and the ordering opposite actual relative mobility at p < 0.0001.

17. We thank an anonymous reviewer for suggesting we examine the spread of estimates and emphasize whether many respondents correctly understand relative mobility, or whether it is only their collective (i.e., average) understanding that is accurate. It is, as Table 5 shows, the latter.