Pregibit: a family of binary choice models
Tóm tắt
The pregibit binary choice model is built on a distribution that allows symmetry or asymmetry and thick tails, thin tails, or no tails. Thus, the model is much more flexible than the traditional binary choice models: pregibit nests logit, approximately nests probit, loglog, cloglog, and gosset models and incorporates the linear probability model. Greater flexibility allows a more accurate estimation of the data-generating process, including asymmetric and thick/thin tails. We prove that the maximum likelihood estimator of the pregibit model is consistent and asymptotically normally distributed. A Monte Carlo analysis and two real-world examples show that probit and logit estimates may show misleading evidence in cases where a pregibit model is statistically preferred. One example concerns enrollment in post-secondary education in Belgium: The pregibit estimate of the enrollment gap between Belgian natives and foreign students is 50 % larger, and the type of high school (general, technical, catholic) is more influential. The second example examines the outcome of mortgage applications in the USA. Here, pregibit estimates assign a stronger role to variables that measure the financial strength of mortgage applicants and a weaker role to demographic characteristics including minority status. More importantly, the distribution of the disturbances proves to be seriously skewed: Pregibit indicates that even high-risk applicants (with a probit acceptance probability of nearly 0) have a positive probability of getting their mortgage application approved. Apparently, mortgage officers are more inclined to uncover reasons to make a mortgage deal than to send clients away empty-handed.
Tài liệu tham khảo
Amemiya T (1973) Regression analysis when the dependent variable is truncated normal. Econom 41:997–1016
Amemiya T (1981) Qualitative response models: a survey. J Econ Lit 19:481–536
Bazán JL, Bolfarine H, Branco M (2010) A framework for skewed-probit links in binary regression. Commun Stat Theory Methods 39:678–697
Freimer M, Mudholkar GS, Kollia G, Lin CT (1988) A study of the generalized Tukey lambda family. Comput Stat Theory Methods 17:3547–3567
Greene WH (2012) Econometric analysis, 7th edn. Prentice Hall, Upper Saddle River, NJ
Hayfield T, Racine JS (2008) Nonparametric econometrics: the np package. J Stat Softw 27(5). http://www.jstatsoft.org/v27/i05/
Joiner BL, Rosenblatt JR (1971) Some properties of the range in samples from Tukey’s symmetric lambda distributions. J Am Stat Assoc 66:394–399
Karian ZA, Dudewicz EJ (2000) Fitting statistical distributions: the generalized lambda distribution and generalized bootstrap methods. CRC Press, Boca Raton
Karian ZA, Dudewicz EJ (2011) Handbook of fitting statistical distributions with R. CRC Press, Boca Raton
Kelchtermans S, Verboven F (2010) Participation and study decisions in a public system of higher education. J Appl Econom 25:355–391
Klein RW, Spady RH (1993) An efficient semiparametric estimator for binary response models. Econometrica 61:387–421
Koenker R, Yoon J (2009) Parametric links for binary choice models: a Fisherian–Bayesian colloquy. J Econom 152:120–130
Liu C (2006) Robit regression: a simple robust alternative to logistic and probit regression. Working Paper, Bell Laboratories
Munnell AH, Tootell GMB, Browne LE (1996) Mortgage lending in Boston: interpreting HMDA data. Am Econ Rev 86:25–53
Nagler J (1994) Scobit: an alternative estimator to logit and probit. Am J Polit Sci 38:230–255
Nelder JA, Mead R (1965) A simplex method for function minimization. Comput J 7:308–313
Pregibon D (1980) Goodness of link tests for generalized linear models. J R Stat Soc Seri C (Appl Stat) 29:15–23
Press WH, Teukolsky SA, Vetterling WT, Flannery BP (1986) Numerical recipes: the art of scientific computing. Cambridge University Press, New York
Ramberg JS, Schmeiser BW (1974) An approximate method for generating asymmetric random variables. Commun ACM 17:78–82
Stukel TA (1998) Generalized logistic models. J Am Stat Assoc 83:425–431
Su S (2007) Numerical maximum likelihood estimation for generalized lambda distributions. Comput Stat Data Anal 51:3983–3998
Tukey JW (1960) The practical relationship between the common transformation of percentages of count and of amount. Technical Report 36, Statistical Techniques Research Group, Princeton University