Statistics Canada's Attention Deficit Disorder

This happens all the time:

With the release of November 2013 data, Statistics Canada converted the Industrial Product Price Index (IPPI) and the Raw Materials Price Index (RMPI) series to 2010=100, with 2010 as the base year. These indexes have also been updated using a weighting pattern based on the 2010 production values of Canadian manufacturers.

At the same time, the classification system was converted to the North American Product Classification System (NAPCS) developed by Canada, the United States and Mexico.

What this self-congratulatory blurb doesn't mention is that Statistics Canada has replaced an IPPI that went back to 1956 with one that goes back to 2010. I can't think of any interesting question that can be answered with a price index that starts in 2010.

Then there's this announcement in a recent Survey of Employment, Payrolls and Hours (SEPH) release:

As of the September 27 release, data in CANSIM tables 281-0023 to 281-0046 cover the period 2001 to present. At the same time, new tables were created for the time series prior to 2001.

Two new tables were created with data starting in 2001: Table 281-0048 (formerly 281-0031 and 281-0034) and Table 281-0063 (formerly 281-0025 and 281-0028). For a concordance vector table, or for more information on these changes, contact Labour Statistics Division Client services.

As of this release, CANSIM tables 281-0050 and 281-0051 on payroll employment from 1991 to 2000 are available (formerly part of tables 281-0023 and 281-0024).

I can't make head nor tail of this. For what appears to be no reason whatsover, StatsCan has taken a data table that went back to 1991 and split it up into two tables that span 1991-2001 and 2001-present. Even worse, the older data has been tossed into the vast and rapidly-expanding swamp of terminated data tables that threatens to swallow the entire Cansim site. A few months ago, someone looking for SEPH wage data would get the whole series. Now, you'll get data going back to 2001 and have to already know (StatsCan won't tell you) that there are older data hidden behind the "Beware of the Leopard" sign.

More examples off the top of my head:

• The series V1996473 (total salaries and wages paid) only goes back to 1997. But it turns out that this is a continuation of series V500265, which runs from 1961-1996. If you stack the two together, you get one long series. StatsCan knows this (they're the ones who told me over the phone), but they always seem surprised to learn that there are people out there who also might want to know how to piece together longer series from the piecemeal snippets in Cansim.
• The national accounts data used to go back to 1961. They stopped those series in mid-2012 and replaced them with data that go back to 1981. I think they're supposed to push back the new data back to 1961 eventually, but it's already been 18 months already. Why wouldn't they wait until they had re-done the entire series and *then* make the switch?
• Same thing for the data on government revenues and expenditures: only 2 tables in this list are being maintained, and you can't even break it out by province. The last year for which you can is 2009. Again, StatsCan is supposed to be working on yet another shiny new data revision and we're all supposed to be very happy with the result. But in the meantime, we might as well pretend that the world ended in 2009 as far as public finances are concerned.
• Then there are the data that StatsCan has and is willing to give out if asked, but only if you happen to know that the data are there in the first place - and they won't tell you what there is. Phil Cross is probably the only person in the world who could have pieced together this business cycle chronology, because as a former StatsCan analyst, he knew that StatsCan had data (in fragments, of course - it wouldn't do for StatsCan to maintain a long  series) for things like unemployment rates going back several decades before the LFS.

I can do this all day.

Statistics Canada must be the only statistical agency in the world where the average length of a data series gets shorter with the passage of time. Its habit of killing off time series, replacing them with new, 'improved' definitions and not revising the old numbers is a continual source of frustration to Canadian macroeconomists.

Update: I'm going to keep track of these episodes as I come across them

• The Survey of Household Spending: It used to go back to 1997, but with the release of the data for 2012, it now only goes back to 2010.

Comments

[Dave Giles]

Stephen - oh yes! I'd love to see their "justifications" for any of this.

[Stephen Gordon]

So would I. I'll share whatever response I get, if there is one.

[Chris S]

"Prices have not changed substantially in the history of price change recording. There is no good reason to adjust tax rate change points based on consumer purchasing power."

"Historically, we have no record that businesses ever paid taxes more than 2% above what they pay now. Any suggestions that businesses used to pay more are not supported by the data."

Paranoid yet?

[Rumplestatskin]

"Statistics Canada must be the only statistical agency in the world where the average length of a data series gets shorter with the passage of time."

Actually, in my experience this is the norm just about everywhere. I'm Australian and out data is much the same - after all, most of the new definitions are agreed by international statistical bodies, and national statistical agencies want to 'keep up' with modern techniques.

It's just that users of the data are often happy to ignore it, or don't spend the time finding out about how the data is collected etc.

[Kailer]

Why can't we just contract this out to Google?

[Warren]

/agree kailer

[Brendon]

So frustrating!!!! I jump at the chance to fill out the survey pop-ups on the StatsCan website just so I can rant at them (I pretend they care) about data revisions, terminated series and the maddening inability to save a data-set for repeated updating. FRED does it so well, why can't they?

[Livio Di Matteo]

Great point Stephen. In the process of blogging, I keep coming across series that end and then restart and it is pretty frustrating.

[Frances Woolley]

Thanks for this post, Stephen. I had no idea some of those short series could be patched together with older ones - even if one does have to go down to the cellar, and look in bottom of a locked filing cabinet stuck in a disused lavatory.

I think there's a fear in the public sector of doing the wrong thing or saying the wrong thing, so people would rather do nothing than do something which might be subject to criticism later. So any series that could possibly be criticized gets buried.

[Stephen Gordon]

I get the impression there are quite a few series like that - at least at the more aggregated levels. Often what happens is there's a change in the way they break things down by industry/occupation/product/etc, but not in the actual total.

[hishamh]

Stephen,

I agree with Rumpelstatskin: this is pretty much standard practice across the world. The biggest reason for making changes is quite simply changing economic structures.

Taking the first example you quoted, the purchasing behaviour of producers have changed, which means a price index based on the old weighting will not be accurate going forward. CPIs are changed based on changing household expenditure patterns, and IPIs are changed based on changing production patterns. GDP base years are changed, because some industries emerge and some disappear.

If you don't make the adjustment, your data will be increasingly unrepresentative of reality. I've noticed that many countries are now standardising to a five year cycle of updating their time series. That actually makes it convenient for people like me who have to deal with cross-country analyses.

From a practical perspective, there are standard methodologies to deal with splicing series from different base years, for example this one from the OECD:

http://www.oecd.org/industry/business-stats/36596185.pdf

They're a bit of work, but it gets the job done, especially if nothing else is available.

[hishamh]

Sorry, I should add this link as well, which has the formulas:

http://www.oecd.org/std/36871378.pdf

[Eigenscape]

The price index change is just "rebasing". It happens all the time to make price index numbers readable (so indexes don't climb to like 5000). No information is loss.

You can get pre-2010 data that is comparable with the new index by dividing the old series by the 2010 index number and combining it with the new series. E.g. if under the old series 2009=280, 2010=300, then under the new series 2009=93.3, 2010=100.

[Stephen Gordon]

Fine, except

1) StatsCan doesn't tell you that. For all anyone knows, these series are completely incompatible.
2) If you look up the IPPI, they don't tell you where the old series are or how/if you can splice them
3) Why don't they do the splicing for us and publish the long series?

[PA]

This is absolutely normal when series are rebased. For example, when the US GDP figures were rebased last summer, all the series apart from the very highest level expenditure components now begin in 1999. The old stuff used to go back to 1947 on a quarterly basis. A couple of years ago the US BLS changed the definition of the average earnings index to all employees from non-supervisory workers. The new series starts in 2010. The BLS is also changing the definition of the US PPI next month.

Just do what everyone else does and splice the old data onto the back of the new data. Either adjust the level if its an index that's been rebased or apply the monthly/quarterly changes in the old series to the back of the new series. There's normally an overlap of a couple of years to check the two series are similar.

[Steve Williamson]

I thought it was only me. My contact with Canadian data comes every three years when I revise the Canadian edition of my textbook. What a pain in the ass! I'm continually splicing time series, or making do with time series that are much shorter than I would like, because StatCan can't be bothered to make their numbers useful. Arrggghh.

[JBélanger]

There seems to be a lot of delay, SC told me the new series on provincial government expenditures and revenues will come out in the next few months for example. But how about coherent series of provincial GDP by industry going back before 1997 ? You can put together two series of nominal provincial GDP and go back to 1984, but there's nothing in real terms. The Conference Board have some going back to 1961...

[Stephen Gordon]

I thought it was only me.

That's probably a widely-held sentiment.

[GS]

Stephen,

Interesting piece, thanks. I'm curious as to whether you find these issues to be a recent phenomenon, or is it a pattern that's always existed with SC? Thanks,

[Stephen Gordon]

This is an old habit, as far as I can tell.

[westslope]

Revising historical data is an expensive undertaking.

StatCan works on cost recovery. The broader market for detailed StatCan socio-economic data wants RECENT data. Please recall that a big chunk of that market believes in things like forecasting (sic) for example.... Please recall that the larger market thinks that a modern economy and policy decisions run on economic impact analysis.

Working Canadian empirical economists need to come up with some clever justification for taking resources from one taxpayer and giving them to another. As special-interest politics seems to be as popular as ever these days, that shouldn't be too difficult.