Tuesday, October 14, 2003

CAUTION: hanging up may be hazardous to your health (or, the unlucky thirteen)

If you can't read Stata output, the output below is lost on you, but here's the gist. In 1975, people in their mid-thirties were asked to participate in the Wisconsin Longitudinal Study telephone survey. Most people said yes, but some refused. Between 1975 and 2000, (age mid-thirties to sixty) some people died. We know they died because either we learned they were dead when we tried to get them to participate in the 1992 round of the WLS, or we found out they were dead through an eagle-eyed search of the Social Security Death Index. Anyway, what the output below shows is that people who refused to participate in the survey are more likely to be dead than people who agreed to participate. Maybe I will use this in my methods class when we talk about causality.
Logit estimates                                   Number of obs   =       9525


------------------------------------------------------------------------------
dead | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
resp75 | -.4144247 .1744284 -2.38 0.018 -.756298 -.0725514
female | -.3612206 .0822002 -4.39 0.000 -.52233 -.2001111
_cons | -2.050844 .1714861 -11.96 0.000 -2.386951 -1.714737
------------------------------------------------------------------------------

Hmm, I wonder if the people who said no to the WLS phone interview in 1992 are more likely to be dead by 2000 than people who said yes. I just ran the numbers, and yep:
Logit estimates                                   Number of obs   =       9076


------------------------------------------------------------------------------
dead | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
resp92p | -.4442695 .211308 -2.10 0.036 -.8584255 -.0301135
female | -.3740024 .125414 -2.98 0.003 -.6198093 -.1281955
_cons | -2.906913 .206363 -14.09 0.000 -3.311377 -2.502449
------------------------------------------------------------------------------

Update, 9:15PM: An obvious possibility is that people who are knock-knock-knocking on heaven's door at the time of the interview are more likely to be refusals. So I looked at whether someone refused to participate in 1975 and threw out everybody who died prior to 1985, meaning that if this is the result of a few people being on their last legs those legs had to last a decade. Interestingly, this hardly affects the magnitude of the RESP75 coefficient at all (below).
Logit estimates                                   Number of obs   =       9385


------------------------------------------------------------------------------
dead | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
resp75 | -.391977 .1964725 -2.00 0.046 -.7770561 -.0068979
female | -.4020621 .0924075 -4.35 0.000 -.5831774 -.2209468
_cons | -2.29816 .1931167 -11.90 0.000 -2.676662 -1.919658
------------------------------------------------------------------------------

Update, Wednesday morning: An e-mail that I just sent elaborating on this association--
> 18 people who were coded as refusals for the 1992 phone interview have

> since died. If they had died at the same rate as the respondents, only
> 5 would be dead. So it must be that there are 13 people out there who
> would be alive today if only they had agreed to be interviewed. I would
> put a little smiley face after that last sentence if I was the sort to
> use smiley faces.
>
> --Jeremy

No comments: