Questions about the interpretation of the xtdescribe documentation

[Wang Xiaobu]

Around page 100 of the official manual xt.pdf (For covenience: xtxtdescribe.pdf), I’m unclear about how to reproduce the output from the second example of xtdescribe. What’s particularly puzzling is that, given the note "idcode*year does not uniquely identify observations", it seems that the command xtset idcode year shouldn't have worked. But my understanding is that the example did successfully run xtset, followed by xtdescribe.



I believe it's very important to clarify the origin of the sample data used in this example and how exactly the output was generated, because the manual appears to be the only known source that explains that the digits in the pattern output can be not only 1 but also 2, 3, and other numbers.

[Chen Samulsion]

That is really strange. Successful xtset panelid timeid requires there to be at most one observation for each combination of identifiers. And the output of xtdescribe is a slightly different from "normal" which gives Delta and Span in separate lines. The following complementary explanation in xtxtdescribe.pdf not clarfying but adding confusion.

In fact, this is a dataset that was itself extracted from the NLSY, in which t is not time but job number. To simplify exposition, we made a simpler dataset by selecting the last job in each year.