Wednesday, April 29, 2009

Biases in Self-Rated Heights


The purpose of this post is to alert researchers towards biases that arise when people self-report their height. I will contrast the way height is reported between two different surveys, the Health Survey for England and The British Household Panel Survey. The 2005 Health Survey for England reports heights objectively – as measured by a visiting nurse using specifically designed apparatus. Wave 14 of the British Household Panel Survey asks participants their heights subjectively. Both surveys specifically want measured height from participants not wearing shoes. In addition, both surveys are nationally representative when compared to the UK census. For the purposes I only look at adults aged over twenty born in England of white race or “British,” ethnicity. Results are split by gender and into year of birth cohorts.

Table 1.






































Table 1 and the accompanying graphs display the degree of biases which take place when people self report their heights. In almost all cases individuals have higher self reported heights in comparison to their subjective heights. There are a number of potential reasons for this. Firstly, people may be rounding their heights. Subjective height from the 2005 BHPS is mostly measured in feet and inches (although participants were given the option to answer in centimeters). The results displayed above are consistent with the hypothesis that people round their heights up.

Examining the frequency distributions of both samples and sexes illustrate this “rounding,” effect. Objectively measured height from the Health Survey for England clearly follows a more normal distribution. The BHPS data is quite staggered and is more discrete than continuous. It also appear that males are more likely to round than females.













The rounding up hypothesis does not explain why the heights diverge for the later year of birth cohorts. This effect seems indicate that the over 60 age category systematically mis-remember their heights. It is widely known and observed that people shrink with age (Diacinti et al. (1995)). The results in table one are consistent with this hypothesis. They also suggest that the divergence between subjective and objective measures of heights are caused by older age cohorts recalling their heights from the early adult years and not taking into account that they may have shrunk over time.

The implications of this post are clear. Self-reported height is biased and researchers should be extremely careful in their analyses and interpretation of it.

Reference:
Diacinti, D., Acca, M., D'Erasmo, E., Tomei, E., Mazzuoli, G.F, “Aging Changes in Vertebral Morphometry,” Calcified Tissue International, 57 (1995), 426—429.

10 comments:

Kevin Denny said...

This is very interesting for those of us interested in anthropometric research. I had always thought there were significant (& possibly non-classical) measurement error issues here. Some people, who shall remain anonymous (like Liam), think they are taller than me when, in fact, they are not.

Martin Ryan said...

Self-reported height is interesting in that it is more concrete than self-reported health, political freedom or skills-matching.

I am thinking of the anchoring vignettes method here; essentially that there is no reason to put forward an "incomparability hypothesis" for self-reported height, in the sense that:

(i) there is no DIF (height is height, after all)
(ii) any problems in measurement vary non-systematically over individuals

What strikes me as the biggest problem is that objective height may be (on-average) over-estimated. But because this is done on-avearge, the consequences may not be an insurmountable concern?

Martin Ryan said...

Alan,

I should acknowledge your statement that the over 60 age category systematically mis-remember their heights; based on the evidence that that people shrink with age (Diacinti et al. (1995)).

However, when stating that self-reported height is biased and researchers should be extremely careful in their analyses and interpretation of it, we should probably emphasise that this is only the case for the over-60's?

Alan Fernihough said...

Martin,

I don't think emphasizing one aspect of this is correct. There appears to be biases running throughout all the year of birth cohorts. The shrinking mis-rememberment hypothesis is just one part of the overall story.

I've also updated some of the graphs on the post now. Distributions are now split by gender with interesting results.

Mark McG said...

I'd like to see the confidence intervals on these

Mark McG said...
This comment has been removed by the author.
Mark McG said...
This comment has been removed by the author.
Mark McG said...

I think it's hard to pin down the extent of "shrinkage" as opposed to sampling error. I've done a quick comparison of heights in SLAN (2002) and Living In Ireland. Figures below are differences (cm) between the two for each cohort. The same "pattern" appears for men, despite the fact that these are both subjective reports.

Year of Birth Males Females
1978-1982 0.29 -0.84
1973-1977 -1.11 -0.39
1968-1972 -1.32 -0.38
1963-1967 -0.17 0.34
1958-1962 0.34 0.53
1953-1957 0.99 0.40
1948-1952 0.93 1.33
1943-1947 -0.29 0.50
1938-1942 -0.02 -0.79
1933-1937 1.36 -0.34
1928-1932 1.68 0.94
1927- 2.00 -0.50

Alan Fernihough said...

Mark,

Your crudely constructed table is very hard to read. However, from what I can see there doesn't appear to be any large differences or trend in either of the gender cohorts. I would also question your smaller sample size and non-random nature of your data sources. I would like to see these two surveys measured against an objective survey of heights amongst the Irish population. I am confident that replicating my analysis on the Irish population would yield very similar results that are consistent with the biases I refer to in my post.

Liam Delaney said...

please lets not get into a debate about shrinking!