Excel Add-In: Gauss Newton Dummy Dependent Variable
Warning: When you download the add-in, make sure that you save it as an ".xla" file. Internet Explorer often changes the file extension to ".xls".
This add-in, DDVGN.xla, estimates Probit and Logit models via non-linear least squares and maximum likelihood. Unlike its cousin, DDV.xla, this add-in does not use Excel's Solver to estimate the coefficients; instead it employs a Gauss-Newton algorithm. As a result, it's quicker, more accurate, and can handle more variables than DDV.xla. In addition this add-in can handle missing values. However, this algorithm does not make clear to the student the structure of the optimization problem. We use DDV.xla to teach and DDVGN.xla to actually estimate models. Remember that for any serious work in econometrics, our results should be checked against those from commercial software.
Download the installation instructions first, then right-click to download the DDVGN.xla file when you are ready. The other files contain documentation and examples. The Probit and Logit NLLS Tests file compares results from this add-in to those from Stata for a data set analyzed by Thomas Mroz.
Latest version: 18 February 2007. Our thanks to Rafael Nicolas Fermin Cota, a student at the Richard Ivey School of Business, Univ. of Western Ontario, for a correction to the algorithm for producing NLLS robust SE estimates.