Having downloaded the data myself, I can only conclude that so far ... they're wrong.
[For those interested in replicating my results, the following is how I did it. Those uninterested may skip this section.]
Downloading the data, (I used the 1972-2006 data set), with the variable AGE (age of respondent), YEAR (year of survey), MARITAL (marital status, 1=Married, 5=Never Married, etc), Sex (1=Male, 2=Female), RACE (1=White), plus of course the always present CASEID, you get a nice text file you can load into MySQL. To make things simpler, I added a constraint in the file created and downloaded, and specified only White Men (SEX=1, RACE=1). Loading this into MySQL gives you interesting results. [I wanted to look at White Men only, avoiding race and class, and looking at what most observers believe the most stable and unchanging group of men in the US.]
One of the things that struck me (and this is why you CANNOT avoid looking at the data in raw form) is how small sample sizes are for White Men in the Age Ranges. For example, the SQL Query
select year, age, count(*) as Num_Age from gss_marital where year = '1975' group by age with rollup;
gives you this output:
|Year||Age||# of Respondents|
The last row is of course the rollup row, showing there were indeed 598 White Men interviewed in 1975. That might seem like a lot, but look at the data at a more atomic level. Only 7 53-year olds, and only 10 54-year olds, in 1975 were interviewed. Your data is only as good as your sample size, and for each age, the samples can be appallingly small. The problem of course is randomness. Can we be sure that for the very small sample sizes for each age, survey takers got a truly random sample of White Men? Much less repeating it for every year? Given that the survey ran from 1972-2006 (I did not use the file with incomplete 2007-2008 data).
select year, count(*) as Num_Men from gss_marital group by year;
select year, count(*) as Never_Married from gss_marital where marital = '5' group by year;
Will each get you the total for each year of all men (all ages), by year, and then the total for all men, all ages, who were never married (MARITAL=5). I've dumped that into an Open Office Spreadsheet to get the following:
[click Image to Enlarge]
Wow. Just for laughs I tried the following:
select year, count(*) as Num_Men from gss_marital where age > '34' and age < '41' group by year;
select year, count(*) as Never_Married from gss_marital where marital = '5' and age > '34' and age < '41' group by year;
And got the following graph (once I loaded it into OpenOffice)
Therefore, it certainly looks as if the data suggests that AT LEAST White Men are getting married later, which would certainly make the "Never Married" status stronger in surveys. The data is not inconsistent with that hypothesis, at any rate. How good is the GSS Data? Not particularly good, given questions about small sample sizes for White Men, at each age, and just how random the selection of the survey takers was, but it is one of the few social surveys we do have covering considerable time periods. We certainly see a fairly consistent rise in "Never Married" over the years, which matches the increased cost for a family, given rising housing prices, and the decline of real wages, in terms of house-buying at least, since the 1970's.
Just as important however, may be the changing expectations of women with respect to marriage. Sandra Tsing Loh, the NPR commentator, writer, and performance artist, has a revealing column in the upcoming Atlantic Montly, which I will be posting on later. It certainly seems among professional urban women, marriage has undergone redefinition. Akin to more of a short-term contract offered to NFL free agents, than anything else.