[Gretl-users] Re: More minor bugs or errors (Riccardo (Jack) Lucchetti) & (Allin Cottrell)

On Fri, 31 May 2019, Summers, Peter wrote: Thanks, Peter. I've found two relevant differences between the Wooldridge and POE4 versions of the Mroz data. First, the POE4 version doesn't contain an "lwage" series, so to do the sorting comparison it's necessary to add it first: series lwage = log(wage) # or similar command Second, in POE the "wage" values for women who are not in the labor force are given as zero, while in Wooldridge they're given as NA. In POE, sorting by wage and lwage are not going to produce the same results at all. Sorting by wage, all the zeros come first. But a zero wage gives an NA log(wage) and NAs get sorted to the end. Hence the completely different results you get from the two regressions in my test script. But this I can't replicate and don't understand. Are you sure? Could you try this variant of Jack's test: <hansl> open <Wooldrige Mroz data> scalar check = min(diff(wage)) >= 0 && min(diff(lwage)) >= 0 printf "check: %d (unsorted: should be 0)\n", check dataset sortby wage scalar check = min(diff(wage)) >= 0 && min(diff(lwage)) >= 0 printf "check: %d (sorted by wage: should be 1)\n", check if check != 1 printf "min(diff(wage)) = %g\n", min(diff(wage)) printf "min(diff(lwage)) = %g\n", min(diff(lwage)) endif </hansl> Allin