MALE - FEMALE WAGE DIFFERENTIALSA. Choice of VariablesB. DataC. National Longitudinal Survey Data (workfile: NLS)1. Some Hypotheses2. Regressions and Results:3. Log-Linear FormulationsD. National Longitudinal Survey Data II (workfile: NLS_SAT)1. Some variables and hypotheses2. Regressions and resultsE. Opportunistic Empirical TestingMALE - FEMALE WAGE DIFFERENTIALS(nlshs72.dta and nlsy.dta)This course is devoted to finding empirical answers to sometimes-controversial issues. Controversy means that there are at least two sides to an issue. Whatever finding you discover, there will always be someone ready to dismiss your findings. You should anticipate all the arguments, counter-arguments, and counters to the counter-arguments. A. Choice of VariablesThis section begins with a discussion on the choice of variables, what kind of biases they might introduce, and how we may want to search for new or different measures to controlfor the potential bias.In trying to detect job market discrimination,1 researchers typically regress the log of wages against characteristics of the wage earner, including gender and/or ethnic background.The dependent variable may itself be biased. For example, if the variable is wage and the main effect of job discrimination is not to hire women in the first place (but once they arehired they get nearly the same wage as men), then fewer women are working and making a wage. The regression results would show little or no discrimination. Fringe benefits are often 25% of salary, but are unlikely to be included in the variable called wage. If fringe benefits do not systematically vary with any of the independent variables (that is, our least squares assumptions are not violated), then there will not be any systematic bias in the estimates of the coefficients (only a greater variability). But if, for example, men are more likely than women to have jobs with extensive fringe benefits, then discrimination against women is likely to be underestimated.Wage or salary is a function of education, the type of industry, marital status and region. The more experience that people have, the more they tend to be paid. But a common proxy variable for experience, age - education - 6, has certain inherent biases. Women tend to be in the labor market for shorter periods and instead specialize in domestic production (having and raising children and taking care of household needs). Hence this variable over estimates the market experience that women have, with a consequent potential for finding discrimination (since it would appear that women make less than their measured experience calls for) even if none existed. Often researchers try to controlfor these biases by discovering whether the woman has children or is married. 1 We are dealing with wage and/or job discrimination, and not other forms of bias that may indirectly effectincome. For example, if young girls are told not to gain market skills, then they will make less because theyhave weaker market skills and not because employers discriminate against them. Our labor market tests are not designed to discover whether the "male dominated culture makes women shy away from gaining these skills"Education is often in linear and quadratic form. Sometimes, people use dummy variablesfor graduation from college, and high school. But the kind of education may be important. If men take engineering and economics and women major in literature, then we would expect men to make more per unit of education. Hence, not controlling for major would find discrimination against women where none existed. (Then again, womenmay not take economics because they know that the financial sector discriminates againstthem, but this would be controlled for and discovered if we had data on major). It would therefore be useful to control for major in college. On the other hand women generally dobetter than men in school, so this aspect is biased in the opposite direction unless we can control for grades.If abilities are positively correlated with schooling, then the estimated marginal effect of schooling (that is, the coefficient of schooling) is upwardly biased since it measures both more schooling and more ability even in the absence of schooling. For our purposes here,that is not a problem since we are looking for discrimination after controlling for schooling and ability.Another determinant of wages is the industry (farming, service, etc.) or the type of job (e.g. clerical, managerial, etc.) the individual has. This however, is a situation where adding a variable may be worse than not including it. To the degree that discrimination prevents a group from obtaining better type of jobs, discrimination is under reported if one controls for job type. The problem is that we do not know whether this is a demand or supply issue. That is whether suppliers of high paying jobs do not like women or women do not like the characteristics of high paying jobs (if it is the latter, not including the variable will introduce bias and make it look like there is discrimination against women where none exists). To illustrate the latter point, consider teaching and secretarial work versus accounting and garbage collecting. Teaching allows a parent of school age children to be with the children after school. To the degree that women specialize in family responsibilities, women may prefer teaching (lower pay but better work schedule) to higher paid jobs as accountants. Women may also prefer working as secretaries over the higher paid but less pleasant job as a garbage collector. (Being a doctor may involve better hours and greater salary, but the investment in human capital is of course much higher.) Thus the decision on whether to include a variable for type of job depends on whether you believe that the supply or demand conditions dominate. One could always try to run the regression both ways. Pay rates may depend on the region, or cost of living. If ethnicity varies greatly by region however, we may need to control for this.In a nutshell, inappropriately including variables or omitting other variables may make it look like there is discrimination against women when there is none or make it look like there is no discrimination against women when discrimination exists. Good econometric studies try to do as good as job as possible so that the results cannot be easily dismissed on these grounds.B. DataSeveral National Longitudinal Surveys have been undertaken. These are