Endnotes

¹ Senior Research Fellow, Philippine Institute for Development Studies (aorbeta@pids.gov.ph). This paper was written while the author was a Visiting Researcher at the Asian Development Bank Institute, Tokyo. Opinions expressed here are solely of the author does not necessarily reflects the view or policies of the Asian Development Bank Institute nor of the Philippine Institute for Development Studies. This paper has benefited from the comments of John Weiss, Haider Khan, and Peter McCawley. Research assistance of Janet Cuenca, Keiko Sasaki, Mihoko Saito, Reiko Nishiura and Nami Sampei are gratefully acknowledged. All errors, however, are solely the responsibility of the author.

² Bollen, Guilkey and Mroz (1995) estimates a very similar problem of the endogeneity of the desired additional children on the contraceptive use equation.

³ Although FIML produces efficient estimates, there are a couple of limitations inherent in the FIML. One, obviously it is much more difficult to estimate although this is increasingly not too much of a concern as most statistical packages can be programmed to generate FIML estimates. In fact there is an available routine in Stata called probitiv that implements FIML estimation of the model above described in Filmer and Lokshin (nd). The second is the natural consequence of system estimation; that is any bias resulting from misspecification in one of the equations is transferred to the whole system. Limited information estimates, such as the two-stage probit, limits the bias to the misspecified equation only.

⁴ Not shown.

⁵ This is from the marginal effects column. Note that probit is a non-linear model so the marginal effects, rather than the coefficients, provide the estimate of the impact of the change in the probability of working to a change in the dependent variable.

⁶ Or about five thousand for a family of five

⁷ There are three things to note in computing impact: (a) this is a tobit model so the marginal effects have to consider censoring, the marginal effects columns computes the unconditional values, (b) the dependent variable is in natural logs so the marginal effects computed are in percentage terms.

⁸ Since the estimation uses deflated values, these are inflated back to the 2002 values using the price index.

Download this Discussion Paper [ PDF 268.9KB| 27 pages ].

Comment(s)

There are [0] comment(s) for this entry. Post a comment.