Jack Welch Tries His Hand at Data Analysis, Sort Of

a | A

By Caroline Baum

After tweeting his way to derision last week, Jack Welch has decided to drop the conspiracy-theory angle and play statistician instead.

You will recall that Welch, the former CEO of General Electric, implied in a Twitter post last Friday that the government had manipulated the data to produce a 7.8 percent unemployment rate in September and help President Barack Obama's bid for re-election. Serious people can debate the statistical anomalies in the monthly data, but they don't accuse career civil servants of corruption.

In an apparent attempt to resuscitate his reputation, Welch took to the op-ed page of the Wall Street Journal Tuesday to support his view that the 0.3 percentage point decline in the unemployment rate in September was "downright implausible." After explaining that he and his wife Suzy have no horse in this race, Welch focused on the hard data:

1) He knows lots of people who are having a hard time finding work.

2) None of Wall Street's finest predicted an unemployment rate below 8 percent.

3) Business people don't believe the economy is growing at the breakneck speed required to lower the unemployment rate by one-half percentage point in two months.

A quick look at recent data reveals that the rate fell 0.5 percentage point between April and June 2010, and 0.7 percentage point between November 2010 and January 2011.

Try as he might, Welch can't let go of his conspiratorial bias, claiming the government's data-collection processes allow plenty of room for human intervention. His one concession? He should have put a question mark after his tweet accusing the government of Chicago-style politics. As if that would have made any difference.

Years ago, I was so upset by reader feedback calling me [unprintable] for my inability to see that the inflation numbers were cooked up in some dark basement somewhere that I decided to do something about it. I got permission from the Bureau of Labor Statistics to spend the day with one of 400 part-time assistants who collect data for the consumer price index each month.

My day with Linda started at a local bakery, where we priced an eight-inch fruit tart; the tart had to be weighed to rule out any skimping on the ingredients. We visited a nursery to check the price on a spathiphyllum lynise in a 10-inch pot. And when Costco had no prunes on the shelf, Linda followed a detailed BLS procedure for introducing a new item to the monthly basket, a process that takes nine months.

Is there noise and volatility in the data? You bet. The BLS reports on the "reliability of the estimates" in a technical note included in each report. There's lots more on the Web. You can read about "sampling errors" and "non-sampling errors"; about confidence intervals; and about the "ARIMA time series" used for the birth-death model. It's a statistician's delight (or nightmare, depending on how you look at it).

Yes, the 873,000 increase in September household employment looks big, especially compared to the 114,000 new non-farm jobs as reported in the monthly survey of establishments. The business survey is more comprehensive and considered more accurate on a monthly basis. Over time the two track one another pretty well.

If you aren't statistically inclined, try common sense to counter the claims that the data are manipulated. If someone were going to cook the numbers in front of an election, couldn't he do better than 1.3 percent GDP growth in the second quarter? Just asking.

(Caroline Baum is a Bloomberg View columnist. Follow her on Twitter.)

Read more breaking commentary from Bloomberg View at the Ticker.

-0- Oct/11/2012 18:58 GMT