3300:
Discrimination PAM in Mortgage Lending: (cont)
Thinking more broadly about the effects of discrimination in the housing market
What do we make of the Munnell et al. evidence if the question is: How does racial discrimination affect the opportunities of blacks and Hispanics to purchase homes in places they would like to live?
Discrimination at other stages of the process
Yinger demonstrates that blacks (with similar characteristics as whites) are shown fewer houses and shown houses in less affluent neighborhoods in audit studies of realtors. Minorities are given less information when they inquire about loans before applying.
What if racial discrimination in other arenas (schooling, the job market, etc.) affects legitimate loan criteria
e.g., affects educational achievement, labor income these variables are used to screen loan applicants controlling for these variables might understate the effects of broader discrimination (this is called over-controlling: I.e., controlling for other outcomes of discrimination that may influence the outcome of interest)
Disparate treatment vs. disparate impact of equal treatment
requiring a 20% down payment may differentially impact minorities due to lower asset accumulation is this a legitimate predictor of loan performance?
PAM 3300:
The Returns to Computer Use Krueger vs. DiNardo & Pischke
Computers and the wage structure Alan Krueger, 1993
The 1980s saw a dramatic increase in wage inequality.
Return to education increased: college graduates earn 34 percent more than HS grads in 1979; 56 percent more in 1991.
One hypothesis for the change: rapid, skill-biased technological change in the 1980s caused changes in the relative productivity of various types of workers.
Kruegers focus is whether employees who use computers at work earn more as a result of applying their computer skills
Proxy for skills: Direct or hands on use of computers:
programming, word processing, e-mail, computer-aided design, etc.
(Krueger) Table 1:
(Krueger) Table 2
should we control for occupation category?
Table 3
Are the effects causal?
The estimated return on computer use declines when more variables are included - suggests there might be important omitted variables. Particular concern is whether highly paid workers are more likely to use a computer b/c of differences in unobserved skills or the nature of their occupations. Krueger addresses in this several ways:
Controls for characteristics of workers - basic human capital and demographic variables; occupation codes; ability (proxied by test scores). While the estimated computer effect is reduced by inclusion of these other variables, it remains positive and statistically significant.
Krueger concludes that the effect represents a causal effect of computer skills on earnings.
Economic theory - compensating differentials: scarce skills (where demand is greater than supply) will command a premium in the labor market Suggests that expansion of computer use may explain 1/3 to 1/2 of the increased return to education Recommends that computer training be part of public and private job training programs.
The Returns to Computer Use Revisited
DiNardo & Pischke (1997) revisit the interpretation of Kruegers estimates Using German data, they document similar wage differentials associated with computer use. German data has more information on the types of tools used in the workplace.
computer calculator telephone pencils work while sitting hand tools (hammer)
The skills associated with these tools (pencils) are clearly not scarce - there should be no compensating differential associated with their use. But the skills may be related to different types of occupations, and the skill requirements in those positions.
DiNardo & Pischke
Table 3
Bottom Line
The fact that we observe similar effects of pencil use suggests that the estimated effect of computer use is not causal. Rather, it is likely to proxy for unobserved attributes of workers and occupations that are more likely to use computers.
Lessons:
Multivariate regression is a fragile tool - in many cases it is not likely to yield unbiased estimates of the treatment effect The key question is whether there are important unobserved factors that are related to both the outcome and the treatment. Understand how Munnell et al. differs from Krueger (and Yule) in this regard!
coming up
Problem Set 3: Due in class Thursday, March 5.
Next time: LaLonde Evaluating the Econometric Evaluations of Training Programs with Experimental Data (linked to syllabus) Then:
Reflection, Review, and Midterm Exam
And Then (A different way forward: Natural experiments): Freedman: Statistical models and Shoe Leather (linked to syllabus)
DiNardo Natural experiments or quasi-experiments (on Blackboard)