Friday, May 01, 2009

Biases in Self-Rated Heights: Part Deux

I am re-posting on the issue of biases in self-rated heights after some skepticism over the content and theories raised in my last post. The graphs above now contain 95% confidence intervals. These results are consistent with the shrinking misrememberment hypothesis.

The biases for the post-1940's year of birth cohorts are not significant. However, the reason I suspect these figures are biased is because people are rounding up by small amounts. Personally speaking I am 5ft 11.5 inches (181.6 cm) although I always round this up to an even 6ft (182.88cm) when asked my height. I propose this measurement error occurs systematically over large sample sizes. For example, if there are 100 5ft 11.5inches in a sample, I would assume that the majority declare themselves to be 6ft based on convenience. Researchers should be wary of the rounding up hypothesis in subjective heights - despite the insignificant results obtained above.


Kevin Denny said...

"misrememberment" sounds very George Bush!
The rounding up hypothesis is rather like the digital preference phenomenon. Ask people how many cigarettes per day they smoke & they don't say 19 or 23:they say 20 or 25. Lisa Farrell has worked on this. Likewise in the PISA data kids give their class size as 30,35 etc.
Looking at the trend upwards in height, I wondered about survivor bias. If the data is based on those who are still alive & short people die younger (height a proxy for health, cognitive abilty..) then the curve will appear steeper than it is.

Liam Delaney said...

Agree with kevin - misrememberment is a bad choice of phrase. Kevin's point about survivor bias is interesting but it doesnt explain Alan's claim that the divergence between subjective and objective heights occurs at later age groups. Alan, correct me if I am wrong but the height divergence for males looks significant at ages way too young to be due to anything like "shrinking".

Michael Daly said...

Interesting thread- Has anyone seen info on self-reported birth weight categories and actual birth weight. I've seen a couple of papers with correlations around .7-.8 which isn't bad considering the room for error in contrast with something like height. These papers don't address where inconsistencies arise in terms of age, gender etc. profile of participants.

Kevin Denny said...

Looking at Alan's graphs I am struck by the increasing height (in the HSE data) of females born in the 1930's. These would have lived through rationing for a significant part of their childhood including puberty. I wonder could this help explain the trend. Of course the same is true for the boys.
Michael: "self reported birth weight"? Hmmm, I didn't think I could read a weighing scales 'till I was a few years old meself.

Alan Fernihough said...

I withdraw the phrase "misrememberment" now!

The graphs and tables suggest the the significant divergence occurs for males in the 1940's birth cohort, making the shrinking group 65+ years of age.