« 150 Years of Canadian National Defence Spending | Main | How to value Aboriginal language television programming »


Feed You can follow this conversation by subscribing to the comment feed for this post.

This is really good stuff, Frances.

Thanks, Steve. Jennifer's piece is worth reading too - there's a lot more to it than I've captured here.

Thanks, Frances!

Two thoughts here, as spokesman for, respectively, the cranky old man constituency, and the "people who worked an apocalyptic shift at our 25th and Oak store on Saturday." (The latter group is more exclusive.)

i) This particular cranky old man is worried about birth rates. If Quebec's policy is working, we need to double down on it. From the logic of the numbers as I follow them, that would not be social welfare-maximising, since it is not a particularly good way of addressing youth or those with marginal attachments to the workforce. Those groups are not very likely to be involved in household formation, up to, and including having babies. Here, I guess I'm just reinforcing the point you've already made. This is a separate, and valuable, policy goal.

ii) My apocalyptic work experience was driven by four(!) new cashiers not showing up for work on Saturday --about a third of scheduled hours. It wasn't even sunny! New cashiers tend to be young and marginally attached, etc. The thing is, part of that marginal attachment comes the fact that they're blowing off shifts. So, sure, transfer payments to these groups may address their problems. In my experience, however, we're shovelling out transfer payments, in the form of student loans. While this is terrible policy, and obviously doesn't help the people who can't manage to enroll in post-secondary education, we can see what the results of more generous transfer payments are likely to be, and, just from a selfish point of view, I'm not sure we want to go further down that road.

If, on the other hand, we could get wage growth happening again (and, in an ideal world, reallocate already-existing social welfare spending on the youth and marginally attached from student loans to something less perverse), we might gain a new perspective on this aspect of the problem. We would probably also see an acceleration in household formation, coming back to the first point.

Erik - I think I'm getting cranky too, because I agree with much of what you say. One quibble, however:

I'm not sure about people who have marginal attachment to the labour force not being involved in household formation. Low fertility is very much an urban phenomenon.

Interesting article, thank you.

It's easy to test the statistical significance of differences across categories with survey data in Stata. See -help svy: tabulate twoway-.

Benoît - thanks, I thought that I tried that I didn't manage to get it to work, but obviously I didn't try hard enough. I'm pretty much a rank amateur when it comes to Stata.

It's not always obvious how to set up the svy command correctly for the PUMFs. it's fairly obvious that one starts off svyset [pweight=WEIGHTVARIABLE] but then I'm never sure how to set up the strata variables. E.g. with the EICS, the only geography variable is six regions, so should one do anything with that? What other clusters etc should one set up?

I would *love* to have someone create a list that describes how to do the svyset command properly for the major PUMFs (LFS, national household survey, CCHS, etc). Interested?

Frances - not really interested, but thank you for the invitation :)

Dealing with Statistics Canada PUMFs can be frustrating. The user guides describe a complex survey design, but users cannot truly account for that design because the relevant variables (stratas, etc.) are not included in the dataset for confidentiality reasons... With the EICS PUMF as with several others, one can only account for unequal selection probabilities by indicating the sampling weight in -svyset-. This yields variance estimates based on linearization methods in Stata, which are approximate (a bit inflated I believe) but acceptable.

Some PUMFs (e.g., GSS) include bootstrap weights to produce design-based variance estimates, and the user guide provides the appropriate syntax for the -svyset- command. The NHS PUMF also includes replicate weights, but in this case Statistics Canada did not bother with Stata's -svyset- and only provided examples of variance estimation with (unecessary complicated) SAS syntax.

We would all benefit if Statistics Canada improved its documentation...

Benoît, thanks for taking the time to write - it's good to know that I'm not the only one who can't figure out how to match the survey design described in the documentation to the Stata PUMF. I am worried that the PUMFs are a low priority these days at Statistics Canada, and that's a real pity.

The comments to this entry are closed.

Search this site

  • Google

Blog powered by Typepad