Changes

=====Issues and Solution===== There are two issues. Why are we using a pcaPCA? Just to get the number of regressors down? The dimensionality isn't that high. And more importantly, one or more PCA components may be picking up a scale effect. We don't want to use the scale regressors in R2 estimation, because they might drive the R2. ''' So we should, for each regression or beforehand, create residuals from estimating the solution is to first regress to estimate the scale effect and then create residuals: reg growthinv17lf growthinv17l numdealsl numstartupsl i.year i.placeid if lowesthighestflag==1 & year>=1995 & year <. predict growthinv17lfres, residuals We then can't use them as nohull in the regress so our dependent variable.list is as follows: quietly capture reg growthinv17lfres nosinglemulti nopair totsinglemulticount totpaircount tothullcount totpairlength tothullarea if placeid==`placeid'& numclusters==`clusters' & lowesthighestflag==1 & year>=1995 & year <. Again using a cut off of 12, there's some slight divergence between the R2 and adjusted R2 maximization points, but not much (2 out of 20). The results looks pretty much like before for the first 20 with some minor differences. 12 out of 20 places have the same answer. The 7 of the 8 remaining are different by just a hull or two. Only Austin is different, with now 21 hulls rather than 11. The R2 adjusted for Austin maximizes at 7. Checking the other cities leads to the following observations:*Portland is still maximized at 4 hulls*Boston maximizes at 16 hulls on R2 and 5 on R2 adjusted. Recall that it was at 11 with N>=11.*New York now maximizes at just 17 hulls, which is a massive drop. But it does look like a clean interior solution.*Palo Alto is down at 12 hulls.*SF is at 21 hulls, way down from its old value in the 50s. *The large set results look more interior and stable than before... the cutoff of 12 looks reasonable too.