Improved Estimates Using Matching Methods, Working Paper.
My paper examines the problem of constructing partial and marginal likelihood approaches to estimating proportional hazards for clustered observations. It turns out that the problem is isomorphic to the problem of relaxing the assumption of independent errors in an extreme-value stochastic utility model. McFadden (D. McFadden 1978, “Modelling the Choice of Residential Location.” Paper No. 25 In Spatial Interaction Theory and Planning Models, edited by Anders Karlqvist, Lars Lundqvist, Folke Snickars, and Jörgen W. Weibull. Amsterdam: North Holland, 75-96) characterizes all discrete choice probability models with univariate extreme value disturbances that are consistent with stochastic utility maximization. He then gives sufficient conditions for the copula to be consistent with stochastic utility maximization. The sufficient conditions describe the set of GEV models. My paper shows that the duration baseline is eliminated from the partial likelihood for clustered proportional hazards when the probability that the first failure is the first observed failure can be modeled as a GEV probability.
“The Location Decisions of Foreign Investors in China: Untangling the Effect of Wages Using a Control Function Approach.” 2010. Review of Economics and Statistics 92(1):160-166.
Xuepeng Liu, Mary Lovely, and Jan Ondrich
Although studies of aggregate investment flows provide
consistent evidence that capital is attracted to low wages, there is almost
no empirical support for this proposition from plant-level location choice
studies. We examine the provincial location choices of firms investing in
“Changes in Rental Housing Discrimination since 1989.” 2008. Cityscape 10(2):301-330.
This study examines the incidence and causes of housing discrimination in qualitative treatment by rental agents, using national audit data sets from the 2000 Housing Discrimination Study (HDS 2000) and the 1989 Housing Discrimination Study (HDS 1989). Using the fixed-effects logit method described by Chamberlain (1980), we control for unobservable factors shared by audit teammates and conduct hypothesis tests for the incidence and causes of discrimination. We find evidence that discrimination is present in HDS 2000 and is caused by both the prejudice of agents and their response to the prejudice of White clients. We also explore changes in discrimination since 1989 and changes in the causes of discrimination since 1989.
As did previous studies of HDS 1989 and HDS 2000, we find that rental housing discrimination against Blacks still exists but also that it declined significantly between 1989 and 2000. These studies indicate that, since 1989, discrimination against Hispanics has not declined as much or as consistently as has discrimination against Blacks. Our new analysis yields several hints about changes in the causes of discrimination between 1989 and 2000. We find a significant increase in discrimination against Blacks by large rental housing agencies and by Hispanic rental agents. We also find significant decreases in discrimination against Hispanics by female agents and that Hispanic renters with children face less discrimination in 2000 than they did in 1989.
Ondrich, Jan, Emily Pas, and John Yinger
Policy makers and scholars have long been interested
attrition, particularly in poor, urban schools. We
the determinants of teacher attrition in five large
areas in upstate New York. We focus on a teacher's
to leave a school district or to leave teaching using
technique for proportional hazards with unobserved
We find that teachers in districts with higher
to nonteaching salaries in the same county are less
leave teaching and that a teacher is less likely to
when he or she teaches in a district near the top of
salary distribution in that county. We also find,
the impact of salary on the probability of leaving
is small and that very large salary increases would be
to offset the impact of concentrated student
the attrition of female teachers.
You See It, Now You Don’t: Why Do Real
Statistics, 85(4): 854-873.
Potential home buyers may initiate contact with a real estate agent by asking to see a particular advertised house. This paper asks whether an agent's response to such a request depends on the race of the buyer or on whether the house is located in an integrated neighborhood. Like previous research about the causes of housing discrimination, this paper uses data from fair housing audits, a matched-pair technique for comparing the treatment of equally qualified black and white home buyers. However, we shift the focus from differences in the treatment of paired buyers to agent decisions concerning an individual house. Using a sample of all houses seen during the 1989 national Housing Discrimination Study, we estimate a random-effect, multinomial logit model to explain a real estate agent's joint decisions concerning whether to show each house to a black auditor and to a white auditor. We find evidence that agents interpret an initial housing request as an indication of a customer's preferences, but also are more likely to withhold a house from all customers when it is in an integrated suburban neighborhood (redlining). Moreover, agents' marketing efforts increase with asking price for white, but not for black, customers; blacks are more likely than whites to see houses in suburban, integrated areas (steering); and the houses agents show are more likely to deviate from the initial request when the customer is black than when the customer is white. These three findings are consistent with the possibility that agents act upon the belief that some types of transactions are relatively unlikely for black customers (statistical discrimination).
Liberalization of Maternity
Ondrich, Jan, Katharina K. Spiess, Qing Yang, and Gert G. Wagner
German federal law has increased the potential duration of maternity leave five times since 1985. A theoretical model demonstrates that the cumulative return probability at potential duration cannot decline unless the mother's employment conditions or career expectations change. We estimate return to work hazards from the German Socio-Economic Panel for women bearing children in the period 1984–1991 and predict cumulative return probabilities for first-time mothers and mothers with a previous birth. The pattern of cumulative return probabilities as potential duration increases is consistent with the hypothesis that employment conditions or career expectations frequently change for mothers taking longer leaves. Copyright Kluwer Academic Publishers.