End of year effect?

I agree with JF who suggested that this chart was mind-boggling.  The chart accompanied a somewhat diffuse NYT article postulating that tax break or shifting medical practice or less apprehension about tired nurses or added labor-inducing stress from visiting relatives may have something to do with more babies being born in December, particularly at month's end.

This chart presumably shows the "spike" in December births, or more precisely, the shift of January births into December.  The trouble with it is its lack of comparability.  We need to compare the 2002-3 trend to some prior year to see the shift.

Even then, we would have seen only one data point.  So it would have been better to plot multiple years.

Finally, after reading the article, I cannot discern the importance of Monday and Friday.  The yellow-pink coloring has not improved my comprehension of the data; it leaves me with more questions than before.

Reference: "To-Do Lists: Wrap Gifts, Have Baby", New York Times, Dec 20 2006.

PS. Please now visit Jon's response.  Kudos for digging out the historical data series and a stellar analysis!

Comments

[Jon Peltier]

There are more in December?
Average 9 days (12/23-12/31): 10300
Average 9 days (1/1-1/9): 10511
Average 10 days (1/1-1/10): 10680
Can't these people do math?

Are the spikes on 12/27 and 12/30 supposed to be significant? You can't tell that unless you have much more than three weeks of data. No values in the three week period fell outside 2 standard deviations of the mean for the three weeks. Shewhart would not find anything unusual in this data.

[Jorge Camoes]

Of course there are some problems with this chart, but the real problem is the data. Imagine that 40 weeks before Christmas there is a general blackout and everyone makes babies that night. There will be a "spike" around Christmas, but does it mean that it is induced by tax cuts? Of course not.

To evaluate if there is a "tax effect" in December, one of the few options we have is to plot the average gestation period by month. That would show us how serious the problem of induced birth is (I guess July would have also a lower-than-average value).

[Jon Peltier]

All that's needed to make sense of the data is to plot more of it, to help separate the holiday effect from the overall day of the week effect. In this page:

http://peltiertech.com/Excel/Commentary/BirthsByDayOfYear.html

I've shown year end data (actually November through January) for a number of years, and I've concluded that there is in fact a year-end holiday effect. However I don't think the births were induced to get a tax break, because the New Year's effect was less than that of Thanksgiving and Christmas, but rather to lessen the inconvenience of having births during holidays. In addition, I show that other holidays have similar if less-pronounced effects.

[derek]

Jon, I'm interested in the data set you got from the NCHS, but after a little searching I've concluded it's either too well hidden, or you went to some trouble to consolidate the years from multiple large PDFs. Would you be willing to share the spreadsheet you wrote?

[Andrew Gelman]

Yeah, this is yet another example of why "Make the time series longer" should be one of the top 10 rules of graphics. And, yes, weekends and holidays are relevant and displayed terribly here.

So many bad things about the graphs, but I particularly like the "Make the time series longer" rule because, when giving graphical advice, people often unnecessarily restrict themselves to working with whatever data happen to be displayed.

Thinking more generally to models as well as graphs: often people don't want to include too long a series because they have a habit of modeling all the data using a single regression. But with the secret weapon, you can have more data and just add to the display without corrupting whatever analysis you've done.

[Jon Peltier]

Derek -

That's right, I spent an hour or so with those lousy PDF files (I suppose I could have written a macro to speed up data fixing in Excel, but I didn't realize how long it would take). The data isn't complete; for most years I only captured the November-January data, and I plotted the only full year I translated.

Send an email and I'll forward you my workbook, such as it is:

jp at peltiertech dot com

[zbicyclist]

If there is an end-of-the-year effect, it may have less to do with taxes and more to do with medical plans. Medical plan coverage often changes January 1 or June 30 in the US.
When our youngest was born, the OB doctors were leaving the HMO and forming their own practice effective the end of the fiscal year. My wife wanted to have them do the (medically required) C-section, but once they left the HMO this would have been very complex. The solution: C-section on the last day of the year, just a couple of days before the "official" due date.
If there are enough examples like this, this would be sufficient to cause an end of year effect.

[Jens]

What is the most interesting about this chart is the difference between highest and lowest data poinst: ca. 13,000 and ca. 6,500. That is actually quite astonishing. Looking somewhat closer, I see that the 'low points' were all public holidays:

25th Dec - Xmas day
29th Dec - a Sunday
1st Jan - hangover day
4th + 5th Jan = weekend days

In my mind, the only thing this chart indicates is that babies don't seem to get born when people don't work...

Kind of makes sense.

Jens

[JF]

Glad you enjoyed the chart!

Jon, yours is much better and more convincing. Any chance you can start designing for the Times?

My other NYT favorite is inevitably in the Business section, there's a stock-price chart with no scale on the Y axis. (Sigh.)

[Jon Peltier]

Jens -

"In my mind, the only thing this chart indicates is that babies don't seem to get born when people don't work..."

This sums it up nicely.

[Kaiser]

Jon, I added a link to your excellent response to this post. Thanks. I also suspect that the effect has much to do with which days of the week the holidays fall each year.