Matt Yglesias

Mysteries of the BLS Seasonal Adjustment Process

Remember on Friday when unemployment rose and 190,000 jobs were lost? Floyd Norris points out that even though these events were widely reported they didn’t actually happen. Instead, according to the Bureau of Labor Statistics the number of jobs went up and the unemployment rate went down. But then a seasonal adjustment factor was applies, and put things into negative territory.

Brad DeLong was inspired to make similar observations in mid-October when the media reported that new unemployment claims had fallen when, in fact, they rose and then were seasonally adjusted into negative territory:

One issue here is that in unusual times the seasonal adjustment process based on the recent past is probably not going to give you the best results. That said, I don’t think dumping the process would help, the labor market really does exhibit a lot of seasonality. But I think responsible journalism should report both numbers fairly prominently. Even if an uptick in economic activity turns out to be purely seasonal, it’s still a real occurrence in the world.

7 Responses to “Mysteries of the BLS Seasonal Adjustment Process”

1. soullite Says:
   November 11th, 2009 at 1:10 pm

   Does anyone think Matt would be saying any of this if GWB was still President?

   No, we’d be talking about reality where jobs really are often seasonal and how you actually need 200k jobs a month just to break even. I know this because I remember 2003.

2. Telling Lies Says:
   November 11th, 2009 at 1:20 pm

   But I think responsible journalism should report both numbers fairly prominently

   You want journalists to report statistics in a full, clear, accurate and informative way. Good luck with that!

3. Rum raisin Says:
   November 11th, 2009 at 1:20 pm

   Does anyone think Matt would be saying any of this if GWB was still President?

   Cobsiderig that the whole team at the Weekly Standard/Corner, Fox News, and Fred Hiatt (of course) would be screaming it to everyone, probably not. It wouldn’t be necessary.

4. Karl Smith Says:
   November 11th, 2009 at 3:00 pm

   But I think responsible journalism should report both numbers fairly prominently. Even if an uptick in economic activity turns out to be purely seasonal, it’s still a real occurrence in the world

   I would strongly disagree with this. There is enough conspiracy mongering and partisan twisting of the data as is. Generally speaking we want a seasonal adjustment. Perhaps — perhaps — there is a reason to discount the seasonal adjustment this time around.

   However, having a general sense that this time is different is not a rigorous empirical technique and thus no responsible authority should be arguing for it. We should acknowledge that no data series is perfect. That in any case your mileage will vary depending on your industry, geography, etc. We further emphasize that data should not “tell you how to feel about your life.” If you’re unemployed that is awful even if the economists tell you the economy is growing. And, then we leave it at that.

   Second guessing the seasonal adjustment with out a very sound reason for doing so is just a data analysis version of “Opinions on the Shape of the Earth Differ.” It is introducing confusion that no rigorous empiricism could support.

5. ostap Says:
   November 11th, 2009 at 3:15 pm

   In other news, pumpkin sales fell 50% on a seasonally adjusted basis in October. Blogger Matthew Yglesias, noting that sales in October increased 700% over sales in September, called pumpkin sellers a misleading bunch of whiners.

6. Dave Says:
   November 11th, 2009 at 3:23 pm

   Eh. I’m a professional economist. For 95% of purposes, the seasonally-adjusted numbers are fine. I sometimes use the non-seasonal numbers, but most people don’t need them. I think including more numbers would just muddy the waters without adding much insight.

7. shah8 Says:
   November 11th, 2009 at 4:03 pm

   Well, one reason that the seasonal adjusted number is acepted is that that the other employment survey had a decline.