1 International Journal of Epidemiology International Epidemiological Association 1997 Vol. 26, No. 4 Printed in Great Britain Socioeconomic Differentials in Misclassification of Height, Weight and Body Mass Index Based on Questionnaire Data GUNNEL BOSTRÖM*, ** AND FINN DIDERICHSEN*, Boström G (Karolinska Institute, Department of Public Health Sciences, Department of Social Medicine, SE Sundbyberg, Sweden) and Diderichsen F. Socioeconomic differentials in misclassification of height, weight and body mass index based on questionnaire data. International Journal of Epidemiology 1997; 26: Background. The purpose of this study was to analyse misclassification of height, weight and body mass index (BMI), derived from mail questionnaires, and its dependency on socioeconomic factors. Methods. A random sample of 4442 adults aged years, living in Stockholm county, Sweden, were in asked about their height and weight. A few months later 3208 of these adults participated in a health examination that included measures of height and weight. In this study we have used occupational class as the socioeconomic indicator. We have used sensitivity as a measure of misclassification of height, weight and BMI. Results. The difference in mean measured height between socioeconomic groups I and III was 2.7 cm for men and 2.0 cm for women. The mean difference in measured height between socioeconomic groups I and III was higher for men than for women in all age groups. The mean difference between self-reported and measured height was 0.6 cm for men and 0.79 cm for women. For weight, the corresponding difference was 0.74 kg for men and 1.64 kg for women. s BMI was more underestimated than men s ( 0.85 for women, 0.40 for men). When using self-reported height and weight for calculating BMI, 81% of the men and 78% of the women were classified correctly, but only 61% of the obese men and 55% of the obese women were identified. The BMI was underestimated in all socioeconomic groups. Manual workers had a lower proportion of underweight, compared to professionals and intermediate non-manual workers, while the objective measure showed the inverse relationship. The prevalence of overweight and obesity in men was 42% higher in socioeconomic group III compared with group I with self-reported data compared with 28% when measured. Underestimation of BMI was highest among women, the obese, the elderly, and male non-manual workers and female manual workers. Conclusions. Our study shows that socioeconomic differences in height, when using self-reported information, involve an underestimation. This means that the height differences between socioeconomic groups in Sweden may actually be higher than that reported by individuals in surveys. The socioeconomic differences in underweight tend to be underestimated for men, as well as obesity for women, when using self-reported information. The socioeconomic differences in overweight and obesity are shown to be overestimated for men. Keywords: height, weight, BMI, socioeconomic differentials, misclassification Data on height and weight from personal interviews, telephone interviews or mail questionnaires are commonly used in epidemiological studies. Such selfreported data with different levels of external control are quite accurate Although several previous studies have shown that for the individual the difference between self-reported and measured height and * Karolinska Institute, Department of Public Health Sciences, Department of Social Medicine, SE Sundbyberg, Sweden. ** The National Board of Health and Welfare, Centre for Epidemiology, SE Stockholm, Sweden. Uppsala University, University hospital, Department of Social Medicine, SE Uppsala, Sweden. 860 weight may be small, such differences affect body mass index distributions (BMI, kg/m 2 ) in various populations. 1,11,12 That members of the upper social classes are taller than members of the lower social classes is a wellestablished finding of social-anthropological research. 13,14 An association between social class and weight has also been shown for women. 15 Previous studies have demonstrated that tall and heavy people underestimate their height and weight, whereas the opposite is true for small and thin people. 1,2,8,11,12 When using self-reported data on height and weight in relation to social class it is possible that this would have an impact on social class differences.
2 SOCIOECONOMIC DIFFERENTIALS IN MISCLASSIFICATION 861 The purpose of this study was to analyse misclassification of height, weight and BMI, derived from mail questionnaires, and its dependency on socioeconomic factors. METHODS In this study, we have used data collected in in four of the 42 primary health care areas in Stockholm county. These four areas represent the areas of 14 primary care teams at five health care centres. The four areas were selected in order to represent different kinds of populations regarding age, social problems and living conditions. A random sample of 6100 inhabitants, aged years, living in these catchment areas first received a questionnaire by mail and were then invited to a health examination. Those who did not respond to the questionnaire were interviewed by telephone. Besides many other questions, the questionnaire included questions about weight and height. The health examination was performed 4 6 months after the questionnaire. It included measuring weight in light street clothes (without jacket, jersey and shoes and recorded to the nearest tenth of a kilo) and height without shoes (recorded to the nearest tenth of a centimetre). The questionnaire had a response rate of 73% out of a total of 4442 subjects. Our study group included 72% of the subjects (3208), for whom we had complete information from the questionnaire and the health examination, as well as information about socioeconomic status. A comparison of the the study group and respondents to the questionnaire revealed that the study group included a higher proportion of women and the elderly (mean age 1.7 years higher). There were no socioeconomic differences in the distribution of overweight and obesity between the study group and all respondents to the questionnaire (Table 1). Socioeconomic groups are defined according to the socioeconomic classification used by Statistics Sweden. 16 This system is based on occupation, and also takes into account educational level of occupation, type of production and position at work. Individuals are classified according to their current occupation. If they were unemployed at the time of the interview, previous occupation was used. Retired people and those over 65 years are included in the analyses. Three socioeconomic groups are defined: manual workers (III), low non-manual workers (II) and professionals and intermediate nonmanual workers (I). In this study entrepreneurs were not included since they constitute a socioeconomically rather heterogeneous group. TABLE 1 Distribution of overweight and obesity by socioeconomic group (I = professionals and intermediate non-manual workers, II = low non-manual workers, III = manual workers) for those who answered the questionnaire and those who answered the questionnaire and participated in the health examination (age-standardized) Socioeconomic groups I II III Questionnaire (n = 4442) Questionnaire and health examination (n = 3208) The studied group was divided into height and weight quartiles. Relative weight was divided into four classes according to cutoff points based on BMI and recommended by the Joint FAO/WHO/UNU Expert Consultations. 17 For men, underweight was defined as 20.1 kg/m 2, acceptable weight as 25.1 kg/m 2, overweight as 30 kg/m 2 and obesity as 30 kg/m 2. For women, the cutoff points were 18.7, 23.9, 28.6 kg/m 2, respectively. As a measure of misclassification of categorial measures of height, weight and BMI based on self-reported data we have used sensitivity with 95% confidence intervals (CI). The bias in the continuous measures of height, weight and BMI are assessed as the mean of the error (self-reported value minus the true value) score. RESULTS The participants in this study were aged years (mean age, 46.0 years for men and 46.5 years for women). The sample consisted of 45% males with a mean measured height of 1.77 m and weight of 77.1 kg, while the females had a mean height of 1.64 m and weight of 64.0 kg. The difference in mean measured height between socioeconomic groups I and III was 2.7 cm for men and 2.0 cm for women (Table 2). The corresponding difference for self-reported height was 2.6 cm for men and 1.6 cm for women. The mean difference in measured height between socioeconomic groups I and III was higher for men than for women in all age groups. After standardization for age the difference in mean measured
3 862 INTERNATIONAL JOURNAL OF EPIDEMIOLOGY TABLE 2 Mean measured height (cm) by gender, age and socioeconomic group and difference in height between manual workers (III) and professionals and intermediate non-manual workers (I) (95% confidence interval, CI) (n = 3208) Socioeconomic Difference 95% CI group mean height for I III III II I I III Upper level Lower level height between socioeconomic groups I and III increased to 3.0 cm for men, and for women the difference decreased to 1.3 cm. Height The mean difference between self-reported and measured height was 0.71 cm (95% CI : ). The mean difference between self-reported and measured height was significantly different between men and women (0.60 for men, 0.79 for women). There was a clear relationship between age and misclassification of height. Overestimation of height increased with age. The elderly, years, overestimated their height in all height quartiles. As height increased, (height quartile), the mean difference between reported and measured height decreased for both genders combined (1.57, 0.84, 0.46, 0.10). Manual workers overestimated their height more than nonmanual workers. When controlling for measured height, the socioeconomic differences decreased. Sensitivity of self-reported height (Table 3) follows the same pattern for men and women, different age groups and socioeconomic groups. The highest sensitivity was found in the highest height quartile and after that in the first quartile. For the elderly, sensitivity in the three first quartiles was lower than for any other group. Weight The mean difference between self-reported and measured weight was 1.24 kg ( 1.36 to 1.12). underestimated their weight significantly more than men ( 1.64 for women, 0.74 for men). underestimated their weight in all weight quartiles, ( 0.41, 0.94, 1.94, 3.30) while men overestimated their weight in the first weight quartile (0.55, 0.21, 0.77, 2.59). For both genders combined, those who were 64 years underestimated their weight significantly more than those who were 64 years. The elderly overestimated their weight in the first weight quartile. Underestimation of weight decreased with age. Younger people in the fourth quartile underestimated their weight by 4 kg. Non-manual workers underestimated their weight more than manual workers. When controlling for measured weight, these differences increased in the first three quartiles. In the fourth weight quartile manual workers underestimated their weight more than nonmanual workers. Sensitivity of self-reported weight (Table 4) follows the same pattern for all groups. The highest sensitivity was found in the first weight quartile and after that in the fourth quartile. Sensitivity was lower for women than for men, except for the first weight quartile. The sensitivity for younger people was lower than for older people, except for the first weight quartile where younger people had the highest sensitivity. The socioeconomic differences were small in the different weight quartiles. Relative Weight (BMI) Body mass index is calculated from height and weight. We know that height and weight distribution differs by
4 SOCIOECONOMIC DIFFERENTIALS IN MISCLASSIFICATION 863 TABLE 3 Number of individuals in measured and self-reported height quartiles by gender. For men quartile 1 = 173 cm, 2 = cm, 3 = cm and 4 = 182 cm. For women quartile 1 = 160 cm, 2 = cm, 3 = cm and 4 = 168 cm. Sensitivity by gender, age and socioeconomic group with 95% confidence interval, CI (n = 3208) Measured height quartile Self-reported height quartile Sensitivity (%) (95% CI) 81 (77 85) 69 (63 74) 79 (75 83) 94 (92 97) Sensitivity (%) (95% CI) 72 (68 77) 70 (65 74) 77 (74 81) 92 (90 95) Sensitivity (%) and 95% CI for age groups years 88 (83 93) 74 (69 80) 79 (74 83) 93 (91 96) years 78 (74 82) 72 (68 77) 81 (77 84) 92 (89 95) years 68 (62 73) 53 (45 61) 68 (60 76) 100 ( a ) Sensitivity (%) and 95% CI for socioeconomic groups III 80 (76 83) 70 (65 75) 76 (72 81) 90 (87 94) II 71 (64 78) 66 (59 74) 77 (71 82) 93 (89 96) I 78 (71 84) 75 (69 81) 82 (78 87) 96 (94 99) a Normal approximation may not be valid in this subgroup. socioeconomic class. We would for this reason expect that BMI calculated from self-reported data would affect socioeconomic groups differently. The mean difference between BMI (kg/m 2 ) calculated from reported and measured data was 0.65 ( 0.70 to 0.60). The BMI was significantly more underestimated among women than among men ( 0.85 for women, 0.40 for men). People who were underweight reported data that resulted in an overestimation of their BMI and this applied to men more than women. The obese gave a more underestimated BMI ( 2.14 for women, 1.71 for men). Underestimation of BMI increased with age, however when controlling the BMI category, this relationship disappeared. People who were underweight overestimated BMI more, except for the elderly, who reported data that corresponded with measured data. Younger, obese people gave an underestimated BMI by 3.11 and the elderly by The BMI was underestimated in all socioeconomic groups. The underestimation of BMI was lower for male manual workers than for male non-manual workers, while the underestimation was higher for female manual workers than for female non-manual workers. After classification of relative weight, as described earlier, into the groups: underweight, acceptable weight, overweight and obesity it was possible to calculate sensitivity for different groups (Table 5). When using self-reported height and weight for calculating BMI, 81% of the men and 78% of the women were classified correctly, whereas only 61% of the obese men and 55% of the obese women were identified. The sensitivity for underweight was only 59% for men and 73% for women. In the age group years the sensitivity for obesity was only 35% (12 of 34 obese people were classified as obese) and for underweight 59%. Low non-manual workers had the lowest
5 864 INTERNATIONAL JOURNAL OF EPIDEMIOLOGY TABLE 4 Number of individuals in measured and self-reported weight (quartiles) by gender. For men quartile 1 = 70 kg, 2 = kg, 3 = kg and 4 = 83 kg. For women quartile 1 = 57 kg, 2 = kg, 3 = kg and 4 = 69 kg. Sensitivity by gender, age and socioeconomic group with 95% confidence interval, CI (n = 3208) Measured weight quartile Self-reported weight quartile Sensitivity (%), 95% CI 87 (84 91) 76 (72 81) 73 (69 78) 82 (78 86) Sensitivity (%), 95% CI 92 (90 95) 65 (60 69) 59 (54 64) 81 (77 85) Sensitivity (%) and 95% CI for age groups years 93 (90 96) 64 (58 70) 58 (51 65) 72 (65 78) years 87 (83 90) 73 (69 78) 67 (63 72) 83 (80 86) years 90 (85 95) 71 (64 78) 71 (64 78) 87 (82 92) Sensitivity (%) and 95% CI for socioeconomic groups III 88 (85 91) 72 (68 77) 65 (60 70) 81 (77 85) II 93 (90 97) 69 (63 76) 65 (58 71) 85 (80 90) I 92 (88 96) 71 (65 77) 67 (60 73) 77 (71 83) sensitivity of the socioeconomic groups (49% for obesity and 59% for underweight). Manual workers had a lower proportion of underweight, compared to professionals and intermediate non-manual workers, while the objective measure showed the inverse relationship. DISCUSSION In this study, the health examination during which height and weight were measured, was performed 4 6 months after the questionnaire. In this time people could increase or decrease their weight. If these changes differ socioeconomically this would influence the results. The present results for gender and age groups, are in accordance with many other studies So far we have not found any studies of misclassification and the consequences on socioeconomic differences. Some studies include education 2,5,7,9 and show that those who are less educated overestimate their height more than those who are better educated. This relationship disappears if height is held constant. In our study the socioeconomic differences for men almost disappear after stratification of height quartiles. For women the socioeconomic differences remain in the first and second quartile. This study has shown that overestimation of height is highest among men, those who are short, the elderly and manual workers. When using self-reported information of height the socioeconomic differences are shown to be underestimated. Apart from hereditary factors, the height to which people grow is also determined by their health and social factors. Differences in average height are therefore sometimes used as indicators of nutrition earlier in life and health problems. A negative correlation has been established almost universally between socioeconomic status (of parents) and the height of children at various ages. 18 In a longitudinal study, where height was measured twice a year, Lindgren 19 followed a group of Swedish urban school children born in , from nine to 17 years of age. Lindgren could not find any clear
6 SOCIOECONOMIC DIFFERENTIALS IN MISCLASSIFICATION 865 TABLE 5 Number of individuals in measured and self-reported body mass index (BMI) categories by gender. Underweight = BMI 20.1 for men and 18.7 for women, acceptable weight = BMI 25.1 for men and 23.9 for women, overweight = BMI 30 for men and 28.6 for women, obesity = BMI 30 for men and 28.6 for women. Sensitivity by gender, age and socioeconomic group with 95% confidence interval, CI (n = 3208) Measured BMI Underweight Acceptable weight Overweight Obesity Self-reported BMI Underweight Acceptable weight Overweight Obesity Sensitivity (%), 95% CI 59 (50 69) 92 (90 94) 72 (68 76) 61 (51 71) Underweight Acceptable weight Overweight Obesity Sensitivity (%), 95% CI 73 (62 83) 91 (89 93) 64 (60 68) 55 (48 62) Sensitivity (%) and 95% CI for age groups years 59 (50 69) 92 (90 94) 61 (53 68) 35 (19 51) years 65 (50 80) 92 (90 94) 72 (68 75) 61 (54 68) years 79 (66 93) 88 (84 92) 64 (58 70) 57 (47 67) Sensitivity (%) and 95% CI for socioeconomic groups III 63 (53 74) 89 (87 92) 70 (65 74) 58 (50 65) II 59 (44 75) 93 (90 95) 67 (61 73) 49 (37 61) I 68 (53 82) 92 (90 95) 67 (61 73) 59 (45 72) height differences between even extreme socioeconomic groups. These findings have been supported by Cernerud 20 and cited as illustrating the success of Swedish policies to establish a class-free society. Nyström and Vågerö 21 have, in their study of adult body height and childhood socioeconomic group in the Swedish population, studied another sample of the same birth cohort in , when they were 25 years old, and found differences in height between higher and lower socioeconomic groups. Their study was based on information collected by interviews. Our study shows that socioeconomic differences in height, when using selfreported information, involve an underestimation. This means that the height differences between socioeconomic groups in Sweden may actually be higher than that reported by individuals in the study of Nyström and Vågerö. Using BMI (based on self-reported data) as a continuous variable would probably have little effect on analyses, but misclassification would occur when using BMI as a categorical variable. Underestimation of BMI was highest among women, the obese, the elderly, male non-manual workers and female manual workers. Socioeconomic differences in underweight tend to be underestimated for men, as well as obesity in women, when using self-reported information. The socioeconomic differences in overweight and obesity tend to be overestimated for men. The prevalence of overweight and obesity in men was 42% higher in socioeconomic group III compared to group I with self-reported data as compared with 28% when measured. REFERENCES 1 Kuskowska-Wolk A, Karlsson P, Stolt M, Rössner S. The predictive validity of body mass index based on self-reported weight and height. Int J Obesity 1989; 13: Stewart A L. The reliability and validity of self-reported weight and height. J Chron Dis 1982; 35:
7 866 INTERNATIONAL JOURNAL OF EPIDEMIOLOGY 3 Wing R R, Epstein L H, Ossip D J, LaPorte R E. Reliability and validity of self-report and observers estimates of relative weight. Addict Behav 1979; 4: Stunkard A J, Albaum J M. The accuracy of self-reported weights. Am J Clin Nutr 1981; 34: Jalkanen L, Tuomilehto J, Tanskanen A, Puska P. Accuracy of self-reported body weight compared to measured body weight. A population survey. Scand J Soc Med 1987; 15: Stewart A W, Jackson R T, Ford M A, Beaglehole R. Underestimation of relative weight by use of self-reported height and weight. Am J Epidemiol 1987; 125: Palta M, Prineas R J, Berman R, Hannan P. Comparison of selfreported and measured height and weight. Am J Epidemiol 1982; 115: Schlichting P, Höilund-Carlsen P F, Quaade F. Comparison of self-reported height and weight with controlled height and weight in women and men. Int J Obesity 1981; 5: Pirie P, Jacobs D, Jeffery R, Hannan P. Distortion in selfreported height and weight data. J Am Diet Assoc 1981; 78: Roberts R J. Can self-reported data accurately describe the prevalence of overweight? Public Health 1995; 109: Kuskowska-Wolk A, Boström G, Rössner S. Influence of body image on estimation of body mass index based on selfreported weight and height. Diabetes Res Clin Pract 1990; 10: Kuskowska-Wolk A, Bergström R, Boström G. Relationship between questionnaire data and medical records of height, weight and body mass index. Int J Obesity 1991; 16: Schumacher A, Knussmann R. Are the differences in stature between social classes a modification or an assortment effect? J Hum Evolut 1979; 8: Macintyre S. A review of the social patterning and significance of measures of height, weight, blood pressure and respiratory function. Soc Sci Med 1988; 27: Ross C E, Mirowsky J. Social epidemiology of overweight: A substantive and methodological investigation. J Health Soc Behav 1983; 24: Socioekonomisk indelning. Statistics, Sweden. Meddelande i Samordningsfrågor 1982:4. (Swedish socio-economic classification, SEI) Reports on Statistical Coordination 1982:4. 17 Report of a Joint FAO/WHO/UNU Expert Consultation(1985) Technical Report Series No 724. Geneva: WHO. 18 Meredith H V. Body size of infants and children around the world in relation to socio-economic status. Adv Child Dev Behav 1984; 18: Lindgren G. Height, weight and menarche in Swedish urban school children in relation to socioeconomic and regional factors. Ann Hum Biol 1976; 6: Cernerud L. Growth and social conditions. Height and weight of Stockholm schoolchildren in a public health context. NHVreport 1991:5. Göteborg: The Nordic School of Public Health. 21 Nyström Peck A M, Vågerö D H. Adult body height and childhood socioeconomic group in the Swedish population. J Epidemiol Community Health 1987; 41: (Revised version received January 1997)