The Predictors of Unemployment (GSS)

Using the GSS data, I try to investigate which factor is the most determinant of the risk of being unemployed, as reported by GSS respondents. For this purpose, I use the logit regression. Below are the variables used in the regression.

UNEMP – At any time during the last ten years, have you been unemployed and looking for work for as long as a month? (0 = YES; 1 = NO)

SEX – “1” corresponds to “MALE” and “2” corresponds to “FEMALE”.
RACE – “1” corresponds to “white”, “2” corresponds to “black” and “3” to “other”.
POLVIEWS – “1” corresponds to “Extremely liberal”, “4” corresponds to “Moderate” and “7” corresponds to “Extremely conservative”.
DEGREE – “0” corresponds to “Less than high school”, “1” corresponds to “High school”, “2” corresponds to “Junior College”, “3” corresponds to “Bachelor” and “4” corresponds to “Graduate”.
WORDSUM – a proxy for IQ, correlation = 0.71; 0.83 for “g”. Recall that blacks do relatively better on vocabulary test (they perform worse on culture-free tests).

Allocation of cases (unweighted)
Valid cases – 13,421
Cases with invalid codes on variables in the analysis – 41,666
Total cases – 55,087

It should be noted that while all the coefficients are statistically significant, the pseudo R-squared, which expresses the proportion of variance in the dependent variable explained by the entire set of independent variables, shows a very low value.

Keep in mind that an independent variable with a low point-scale (say, 2) is expected to have a higher coefficient than an independent variable with a high point-scale (say, 10). In fact, a change in one unit of an independent variable with a high point-scale would have a very little effect, especially when the independent variable can take on many values (for instance, years, age, or income), on the dependent variable.

Unsurprisingsly, DEGREE and WORDSUM have positive Bs, which means that intelligence and degree level are negatively (and independently) associated with the risk of being unemployed. RACE has a negative B, which means that being black (or other), regardless of degree level and intelligence, is significantly associated with the risk of being unemployed. One may be tempted to jump to the conclusion that discrimination and racism account for the racial differences in unemployment rates, but evidence shows that the hypothesis is false. In contrast, the minimum wage is the best explanation of this racial disparity.

A curiosity is the fact that POLVIEWS is strongly associated with the risk of being unemployed. Why are liberals more likely than conservatives to be unemployed ? And why POLVIEWS is more determinant than DEGREE and WORDSUM in the likelihood of being unemployed ? (An anonymous referee points out to me that the causation could go the other way, that is, unemployed tend to become liberal, relying on government support.)

Then, I re-ran a logit regression, including “SEX” in lieu and place of “RACE” and by limiting the sample to whites (“RACE(1)” in the filter).

Allocation of cases (unweighted)
Valid cases – 11,342
Cases excluded by filter or weight – 10,214
Cases with invalid codes on variables in the analysis – 33,531
Total cases – 55,087

As can be seen, the coefficients are statistically significant, although WORDSUM is statistically significant only if we accept a 10 percent level of significance (p-value < 0.10).

Here, another curiosity emerges. Why being a male is so strongly associated with the risk of being unemployed ? Note that this result is also at odds with the feminist mantra that women are oppressed and discriminated. I wonder whether affirmative action may account for this gender difference.

Finally, I investigate whether POLVIEWS affects the risk of being unemployed by including the variable “AGE” since unemployment rate for young people is usually high, thanks to minimum wage laws, and because young people are also more likely than old people to be a liberal. To control for this potential confounding factor, I include “AGE” and “COHORT” in the regression. Because the GSS asks the question in every survey year since 1973, we should also control for the survey year. COHORT may be used for this purpose.

COHORT (r:1883-1910;1911-1930;1931-1950;1951-1970;1971-1992). 5 point-scale. Birth cohort of respondent.
AGE (r:18-25;26-33;34-41;42-49;50-57;58-65;66-73;74-81;82-99). 9 point-scale. Respondent’s age.

Allocation of cases (unweighted)
Valid cases – 13,407
Cases with invalid codes on variables in the analysis – 41,680
Total cases – 55,087

We can see that AGE has a positive B, which means that unemployment decreases with age, consistent with the finding that young people are less productive, and have fewer skills than older people. The negative sign of the coefficient for COHORT underlines an increasing rate of unemployment over the years under study. Using YEAR variable – recoded as follows : YEAR (r:1972-1980;1981-1989;1990-1999;2000-*) – in place of COHORT yields the same result. But more strikingly is the fact that, even when AGE and COHORT have been controlled for, being a liberal still remains associated with the risk of being unemployed, for some obscure reasons. It is also worth noting that being a black (or other) is still associated with the risk of being unemployed. Another factor beyond age and intelligence must account for the racial difference in unemployment rate. Again, minimum wage is a good candidate.

About these ads

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s