Marc Novicoff and five other group members--names redacted in case they do not want this published under their names
Econ 20 Term Project
Dumb and Dumber:
A Study of the Wage Effects of Being the “Dumbest” Sibling in Your Family, Even if You’re Smart
Abstract
Our project looks at the wage effects of being the dumbest sibling in a family. We take “Y” as the log of the wages in order to examine percent change in wages that comes from being the dumbest sibling in a family. Our “X’s” include the dummy variable of an individual being the dumbest sibling in a family based on a standardized 1979 army intelligence test, as well as several controls — binned intelligence to make sure we aren’t just regressing wages on intelligence, and dummies for Hispanic, Black, male, and urban/rural. Our data was extracted from the 1979 National Longitudinal Survey of Youth (we dropped other years), which gave us a cross-sectional dataset. While we hypothesized that siblings who were the dumbest in their family would be worse off than similarly intelligent siblings who were not the dumbest of their family (due to low self-esteem for example), the opposite proved true; we found that dumbest siblings earned roughly 5 percent more than their similarly intelligent peers who had higher family rank (with statistical significance at the 10% level). There are several possible reasons, including a “networking” effect where your connections with smarter people — in this case, with smarter siblings and probably smarter parents— leads to better job opportunities. There is also the possibility that those with lower family rank have a better work ethic to compensate for their low family rank.
Introduction
The 1979 National Longitudinal Survey of Youth offers a convenient opportunity to study the relationship between AFQTRev (Armed Forces Qualification Test: revised version) scores and future earnings. The data sorted at a household level allows us to classify a sibling as the “dumbest” if he or she received a lower AFQT score than any of his/her siblings did.
Our hypothesis was that being the dumbest sibling in your family (as measured by the 1979 revised armed forces qualification test) is associated with lower earnings. We believed that being the dumbest sibling in your family causes you to feel worse about yourself, and that negative self-esteem would make you a less successful (and money-making) person. We used the National Longitudinal Survey of Youth in 1979 report for data. Our X variable is essentially the dummy variable “dumbest” to represent whether or not a person is the dumbest sibling. We also controlled for intelligence (in 20 bins so we have more data per bin) so we are not just measuring the effect of intelligence on wages. In addition, we controlled for Hispanic, Black, male, and urban/rural. Our Y variable is log wages in 1989. The coefficient of interest will be on “dumbest,” which will estimate the effects of being the least intelligent child in the family on log wage. We used these variables to estimate the effects of being the dumbest sibling in the family on the wages in an OLS regression.
The assumption we need for causality is that being the least intelligent sibling (as measured by the army’s test) in any given family is random. The fundamental assumption of causal inference for our study is that: if the “dumbest” sibling in the family were NOT the “dumbest” sibling, they would have had the same earnings outcome as people who are not the dumbest sibling. We are also assuming the third factors affecting earnings are uncorrelated with being the dumbest.
There are many sources of omitted variables bias. For example, race, gender could be hidden factors that limit income or employment status. We are going to add these as controls in our multivariate regressions. Other sources of omitted variable bias might be work ethic or a measure of how good a match you are for your specific job, rather than just your cognitive ability.
In our study, we found a positive relationship between being the “dumbest” sibling and ln(wages). Specifically, when we regressed ln(wage) on the dummy for being the “dumbest” sibling and dropped “only siblings”, we found a slope coefficient of .049, significant at the 10 percent level (Table 3, Column 4). This means that holding constant controls for race, gender, and urban, being the dumbest sibling is associated with 4.90 percent higher wages. While finding this positive relationship between being the “dumbest” sibling and future wages appears counterintuitive, these results may be driven by the aforementioned omitted variables, such as work ethic or networking.
Data and Methods
f(Afqtrev) is a collection of twenty dummy variables that indicates which of the 20 ventile groups of intelligence scores an individual is in.
Our data comes from the 1979 National Longitudinal Survey of Youth: a “nationally representative sample of 12,686 young men and women who were 14-22 years old when they were first surveyed in 1979. These individuals were interviewed annually through 1994.” For our purposes, we wanted a cross-section of data, so we dropped all the data that wasn’t in the year 1989.
The key regression we are running is the natural log of wage on our dummy variable “dumbest,” signifying whether a person is the least intelligent sibling in their family as measured by the army (1 if least intelligent, 0 if not or only child), with controls for being black, being Hispanic, a control for living in an urban area or not, and industry dummies. In addition, we have controlled for intelligence so that we are not simply doing a regression on the returns to intelligence, but rather measuring the returns to having the lowest family rank. This control for intelligence is done by using twenty bins of intelligence, under the assumption that one’s very specific intelligence score is not so accurate and probably not relevant, but which ventile they’re in probably is.
Results
Table 1: Table of Means
Legend
lnwage = the log of the wages
Dumbest = 1 if the child is the dumbest in a family
Afqtrev = the percentile which the subject scored on the revised AFQT intelligence test in 1979.
Hispanic, Black, Other = race dummies
Urbrur = 1 for urban, 0 for rural.
Table 2 : Balance Test
(Note: The highly significant relationship between being male and being “the dumbest sibling” may be because AFQT tests are correlated with education and many females attain more education, possibly because females feel they must do well in school and get well-educated to make up for gender discrimination in the workplace. There are other reasons this may be true, we are sure, but we controlled for male for this exact reason.)
Table 3: OLS Effects on Wage Estimate
(click to expand and see clearly)
Table Notes:
Legend: urbrur=1 if urban, 0 if rural
Controls: Hispanic, Black, Male, Urban
Standard errors were clustered on household ID.
Afqtrev bins are split into 20 bins (ventiles) for regressions 2-4. For regression 1, we control for afqtrev, but not in bins
Regression interpretations:
The regression 4 (key regression) estimate is:
In regression 4, excluding only children:
𝜷0 = 9.499 | 𝜷1 = 0.049 | 𝜷2 = 0.092 | 𝜷3 = 0.030 | 𝜷4 = 0.513 | 𝜷5 = 0.154
Interpreting the slope on dumbest, having the lowest intelligence of any sibling (while not being an only child), is associated with a 4.9% higher wage, which is statistically significant at the 10% level, after controlling for dummies relating to Hispanic, Black, male, and urban, as well as excluding all only children in each family.
Figure 1: ln(wage) by AFQTREV
In our graph, the dumbest siblings’ performance is denoted by the blue dots and red best fit line, while the other siblings’ performance is tracked with green dots and orange best fit line. We split the siblings into bins of 20 ventiles. The dumbest siblings at higher ventiles of Afqtrev eventually do stop outperforming their higher rank peers, as the slopes intersect and the higher rank siblings pass the lower rank siblings, but at lower Afqtrev scores, low intelligence rank is associated with higher wages.
Discussion and Conclusion:
We are getting a statistically significant effect of being the dumbest sibling having a larger wage on average for most ventiles, which is contrary to our initial hypothesis that being the dumbest sibling would lead to a lower wage. Therefore, we have come to the conclusion that it is actually more beneficial to be the dumbest sibling in the family than the smartest sibling in the family (assuming the same intelligence). Controlling for race and sex decreased the statistical significance of the regression, but it is still statistically significant at the 10 percent level. We think possible sources of omitted variables bias is industry, as industry is likely correlated to wages, and even correlated to dumbest (some industries might have more lower-ability jobs). Another possible form of omitted variable bias may come from work ethic. This in a way would be sort of causal: lower family rank leads to higher work ethic leads to higher wages. However, if the relationship between “dumbest” and wages were via the networking effect (we think this is the most likely), it wouldn’t really be causal so much as the higher intelligence of siblings, and probably parents, would be. Additionally, it is possible that education level may have an omitted variable because AFQT scores are likely correlated with education levels according to the military and education is also correlated with higher wages. Because there are some key omitted variables, we aren’t ready to cause this causal, nor could it ever really be directly causal--it’s clear that low family rank doesn’t literally cause wages; that would make no sense. It may be indirectly causal (through work ethic due to low family rank or networking with smarter siblings), but direct causality seems impossible. It also seems possible that there is just no causality though, and significant variation is due to the OVB or chance.
The fact that lowest-intelligence siblings are earning more than people with similar intelligence who have higher rank in their family is surprising. This leads us to the conclusion that there is probably some “networking” effect, where having smart siblings (and likely smart parents since intelligence is at least somewhat hereditary) is useful in the job market since they can be successful and bring you along or connect you to other successful people. Other more ideal strategies we can use to explore the same question are to run the same study again using a score other than an AFQT score. AFQT scores have been shown to be highly correlated with education rather than intelligence. They are even more correlated with education than a typical IQ test. A test that gauges intelligence more accurately could possibly yield a different result. Our research suggests that the “networking effect” has a significant impact on future earnings, since the “dumbest" sibling in a family achieved higher future earnings. In order to level the playing field for those without familial networking opportunities, the government may consider providing networking opportunities to all students at universities or through a mentor matching program.