Monday, March 10, 2008

"The good psychic would pick up the phone before it rang. Of course it is possible there was noone on the other line. Once she said "God Bless you" I said, "I didn't sneeze" She looked deep into my eyes and said, "You will, eventually." And damn it if she wasn't right. Two days later I sneezed." - Ellen DeGeneres

***

Various sources on what to do when you have more than one endogenous variable:


"When more than one endogenous variable appears on the right-hand side of the equation, the linear combination for each is adjusted so that it will be orthogonal to the first stage residuals for all of the endogenous variables in order to insure consistency."

"The method of estimation used for equations containing more than one endogenous variable is limited information maximum likelihood... The constant reiteration of the word biased when referring to least squares is, however, irritating. Since when has limited information been unbiased ?"

"Whatever omitted variables are excluded from the equations are presumed to affect more than one endogenous variable in the system; this is the seemingly unrelated regressions model"

"Several cross-industry studies model the determination of more than one endogenous variable using a simultaneous equations approach. Schmalensee argues that even these models are unlikely to provide consistent estimates of structural parameters."

"In the case of more than one endogenous variable, Staiger and Stock (1997) show that while it is not possible to evaluate the relative bias of IV directly, it is possible to place an upper bound on it. Their “worst case” relative bias measure... is equal to the inverse of the minimum squared sample correlation between the endogenous variables and the instruments after the exogenous covariates have been partialled out, multiplied by the ratio of instruments to observations"

"Where there is more than one endogenous variable, seemingly unrelated regression equations (SUR) or structural equations modeling, described in Section 3, are more appropriate."

"Although one right-hand side endogenous variable is the most commonly occurring situation (cf. Hanh and Hausman, 2003), applications may suffer from two or more endogenous regressors. For instance, marketing managers not only set prices based on unobserved information, but also other marketing mix variables like advertising or shelf-space location (Chintagunta, Kadiyali, and Vilcassim, 2003, Manchanda, Rossi, and Chintagunta, 2004). Furthermore, in estimating the return to schooling it is common to include measures for experience and squared experience that are constructed from ‘years of schooling’, and hence also endogenous (Verbeek, 2000). The nonparametric Bayes approach in chapter 7 is applicable to problems with more than one endogenous variable."