Statistical nerds, your nation needs your help!

April 19th, 2016 at 12:57 pm

As OTE’ers know, I have long lobbied for being as explicit as possible about the uncertainty in our economic estimates and forecasts. Which is why I was very happy to get this survey from the Bureau of Labor Statistics today, asking whether they should present the monthly payroll jobs number (the net gain/loss of payroll employment) with confidence intervals. Right now, you just get the point estimate, with no information as to, say, the 90% confidence interval, i.e., the range of values around the monthly estimate for which there’s a 90% chance that the actual value is within that range.

For example, showing this interval would allow you to see whether the confidence interval around the monthly change “crosses zero,” meaning the actual change that month could have been negative or positive (the 90% interval around the overall payroll change is about plus or minus 100,000 jobs).

I would never try to influence any votes, of course, but I think this practice would constitute a real advance. The extent to which noise moves markets is…um…pretty unsettling. Of course, this won’t change that, but it’s a small step the right direction.

The Bureau offers two choices as to how to present the material. Again, the main thing is you take the survey, but I found that I had to do mental math on chart 2 to get back to the result in chart 1. Just sayin’…

BTW, don’t ya just love BLS? They’re not only constantly putting out great stuff. They’re trying to figure out how to make it better, and they’re reaching out to us users for input. Warms me wonkish heart.

Print Friendly, PDF & Email

8 comments in reply to "Statistical nerds, your nation needs your help!"

  1. Chuck Sheketoff says:

    Thanks for the post…and yes, love BLS.

  2. Jill SH says:

    Yes, I like the idea of showing the confidence interval. (Can just anyone take that survey — like semi-retired graphic designers who consider themselves lay policy wonks? Don’t want to skew results if the survey is really intended for pros like you.)

    Something that I think is also important is showing how accurate past projections are to real-world results. (That’s what that Bank of England chart shows, right?) If history shows accuracy on past projections, that will boost confidence in current projections. Or, of course, if not accurate, a learning opportunity!

    I bet, though, that the mainstream media will only report THE jobs number and won’t know what, if anything, to say about intervals.

    I veered away from a life in numbers when I dropped my econ major nearly 50 years ago. But I sure do still appreciate all you number crunchers out there keeping the reality-based world alive.

  3. Chris G says:

    I know it’s supposed to be a “There is no right or wrong answer.” type of question but there’s a correct and an incorrect answer in choosing between between Chart 1 and Chart 2. Chart 1 is the correct choice. Who on Earth displays confidence intervals the way they’re shown on Chart 2? No, no, no. (Sorry, but it needs to be said.)

    And a quibble: The text above the chart reads “Since the range includes values less than zero, we could not say with confidence that construction employment had, in fact, increased for the month.” That’s very sloppy language. You can say that the probability that employment increased is [X]. Why not just leave it at that? What’s special about 90% confidence? Or 95%? Or 82.374%? The data are what they are. Report the summary characteristics of the probability distribution function and let the reader decide what they’re (subjectively) confident about.

    • Chris G says:

      Re-reading my comment, it misses the point that I think adding confidence intervals is a great idea.

      I recall your interview with [Northwestern prof? name escapes me at the moment] where he mentioned that reporting uncertainties was standard practice in the UK but there was no stomach for in DC. That latter didn’t surprise me but it was very disappointing nonetheless. When someone reports a number it’s important to understand the limits of interpretation.

  4. Tom in MN says:

    So do you think this would help the fight against those that claim unemployment is a greater by a large factor than is what is reported? Sure us nerds would like the interval to be given, but aren’t your smoother plots better than adding more numbers for people who think half of the people are above average?

  5. Radhika says:

    I really do love the BLS….what are your thoughts on the manipulation done in such kind of surveys…