- Top of page
- Research Methods and Procedures
Objective: To identify the determinants of underreporting BMI and to evaluate the possibilities of using self-reported data for valid obesity prevalence rate estimations.
Research Methods and Procedures: A cross-sectional monitoring health survey was carried out between 1998 and 2002, and a review of published studies was performed. A total of 1809 men and 1882 women ages 20 to 59 years from The Netherlands were included. Body weight and height were reported and measured. Equations were calculated to estimate individuals’ BMI from reported data. These equations and equations from published studies were applied to the present data to evaluate whether using these equations led to valid estimations of the obesity prevalence rate. Also, size of underestimation of obesity prevalence rate was compared between studies.
Results: The prevalence of obesity was underestimated by 26.1% and 30.0% among men and women, respectively, when based on reported data. The most important determinant of underreporting BMI was a high BMI. When equations to calculate individuals’ BMI from reported data were used, the obesity prevalence rate was still underestimated by 12.9% and 8.1% of the “true” obesity prevalence rate among men and women, respectively. The degree of underestimating the obesity prevalence was inconsistent across studies. Applying equations from published studies to the present data led to estimations of the obesity prevalence varying from a 7% overestimation to a 74% underestimation.
Discussion: Valuable efforts for monitoring and evaluating prevention and treatment studies require direct measurements of body weight and height.
- Top of page
- Research Methods and Procedures
The prevalence of obesity is increasing in many parts of the world, and its impact on the public health and on individuals is widely accepted (1, 2). Currently, a large number of epidemiological studies are being published in which prevalence estimations of obesity are reported and in which obesity is linked to several health outcomes. Large-scale studies regarding BMI levels are often based on self-reported body weight and height. Use of self-reported weight and height could lead to underestimations of obesity prevalence rates because subjects tend to underreport their body weight (3, 4, 5, 6, 7, 8) [especially the obese (9, 10, 11, 12, 13, 14, 15, 16, 17)], and subjects tend to overreport body height (4, 5, 6, 7, 8, 9, 10, 11, 18, 19). However, the severity of underreporting has been questioned (19, 20, 21), and some authors argue that mean levels of BMI may be estimated relatively well by use of self-reported data.
The most important determinant of underreporting body weight seems to be a high true body weight, and subgroups have been mentioned in the literature that underreport weight or overreport height more than others. High educational level, older age (5, 9, 10, 19), smoking (14, 22), female gender (6, 19), diabetic status (22), and a digit preference (9, 15), i.e., rounding off to values ending with 0 or 5, have been identified as potential determinants for underreporting body weight, although some studies find opposite associations.
In a few studies, linear regression equations were calculated by which true obesity prevalence rates could be estimated from reported body weight and height (9, 11, 12, 13, 22, 23). Some of these linear regression equations were concluded to be valid within the sample from which the linear regression equation was formulated (11, 13, 22). One study [National Health and Nutrition Examination Study II (NHANES II)] developed linear regression equations from one-half of the sample to test whether the equations were valid in the other one-half of the sample and concluded that self-reported BMI is difficult or impossible to correct by the use of such equations (23). To the best of our knowledge, no attempts have been made to evaluate the validity of linear regression equations across populations. The present study aims to identify the determinants of underreporting body weight and BMI and to evaluate the possibilities of correcting reported body weight and height, based on both the present and published studies.
- Top of page
- Research Methods and Procedures
Table 1 shows that mean body weight was 1.0 and 1.4 kg lower when based on reported data than when based on measured data among men and women, respectively. Most subjects reported a body weight within 10% of their measured body weight (∼97% of the men and 95% of the women). Approximately 80% of the men and 77% of the women reported a body weight within 5% of their measured body weight (data not shown). Mean body height was 0.5 and 0.6 cm higher and mean BMI was 0.4 and 0.6 kg/m2 lower among men and women, respectively, when based on self-reported data (Table 1). Correlation coefficients for measured and reported values of body weight, height, and BMI were 0.96, 0.95, and 0.93, respectively, among men and 0.97, 0.94, and 0.95 among women. Among obese men, correlation coefficients for measured and reported body weight, height, and BMI were 0.91, 0.96, and 0.76, respectively; among obese women, the correlation coefficients were 0.93, 0.92, and 0.88, respectively.
Table 1. Means and prevalence rates (with 95% confidence limits) among Dutch men and women, ages 20 to 59 years, who had both measured and reported data in the period 1998 to 2002
|Men (n = 1809)|| || || || |
| Weight (kg)||83.9||84.5 (83.9 to 85.1)||83.5 (83.0 to 84.1)||1.0 (0.2; 1.8)|
| Height (cm)||181.4||180.8 (180.5 to 181.2)||181.3 (181.0 to 181.6)||−0.5 (−1.0; 0.0)|
| BMI (kg/m2)||25.5||25.8 (25.6 to 26.0)||25.4 (25.2 to 25.6)||0.4 (0.2; 0.6)|
| Overweight (%)||54.0||58.0 (55.8 to 60.3)||51.8 (49.5 to 54.1)||6.2 (4.6; 7.8)|
| Moderate overweight (%)||43.5||46.6 (44.2 to 48.8)||43.3 (41.0 to 45.6)||3.3 (1.4; 5.2)|
| Obesity (%)||10.4||11.5 (10.0 to 13.0)||8.5 (7.2 to 9.8)||3.0 (2.0; 4.0)|
|Women (n = 1882)|| || || || |
| Weight (kg)||69.6||70.1 (69.5 to 70.7)||68.7 (68.2 to 69.3)||1.4 (0.6; 2.2)|
| Height (cm)||168.2||167.8 (167.5 to 168.1)||168.4 (1.68.0 to 168.7)||−0.6 (−1.0; −0.2)|
| BMI (kg/m2)||24.6||24.9 (24.7 to 25.1)||24.3 (24.1 to 24.4)||0.6 (0.3; 0.9)|
| Overweight (%)||38.6||41.5 (39.3 to 43.7)||35.5 (33.3 to 37.7)||6.0 (4.6; 7.4)|
| Moderate overweight (%)||28.5||30.5 (28.4 to 32.5)||27.8 (25.8 to 29.8)||2.7 (1.0; 4.3)|
| Obesity (%)||10.1||11.0 (9.6 to 12.5)||7.7 (6.5 to 8.9)||3.3 (2.5; 4.2)|
Both the prevalence of overweight and obesity were underestimated when based on self-reported body weight and height (Table 1). The prevalence of obesity was 3.0% and 3.3% lower among men and women, respectively, when based on self-reported data. Thus, as a percentage of the measured prevalence of obesity, obesity was underestimated by 26.1% among men and by 30.0% among women when based on reported data (Table 1).
Underreporting of body weight varied more with measured body weight and BMI than with measured body height, age, educational category, smoking, or reporting weight with a digit preference (data not shown). Underreporting body weight showed a clear dose-response relation with larger values of measured body weight and BMI. Men with body weight < 75 kg or BMI < 23 kg/m2 and women with body weight < 60 kg or BMI < 20 kg/m2 overreported body weight. Among obese men and women, mean body weight was 3.9 and 4.2 kg lower, respectively, when based on reported data than when based on measured data (Table 2). Small body height in women and old age were slightly associated with underreporting body weight and BMI. Educational level and smoking were not clearly related to underreporting. A digit preference was associated with underreporting among women, but not among men (data not shown).
Table 2. Means and prevalence rates (with 95% confidence limits) among obese* subjects
|Obese men (n = 182)|| || || |
| Body weight (kg)||106.3 (104.7; 107.9)||102.4 (100.0; 103.9)||3.9 (1.7; 6.1)|
| Body height (cm)||180.3 (179.3; 181.3)||181.0 (179.9; 182.1)||−0.7 (−2.2; 0.8)|
| BMI (kg/m2)||32.7 (32.4; 33.0)||31.2 (30.9; 31.5)||1.5 (1.0; 2.0)|
| Obesity (%)||All||65.9 (59.4; 72.3)||34.1 (27.7; 40.6)|
|Obese women (n = 196)|| || || |
| Body weight (kg)||93.5 (91.7; 95.3)||89.3 (87.6; 91.0)||4.2 (1.8; 6.6)|
| Body height (cm)||166.4 (165.6; 167.2)||167.4 (166.5; 167.2)||−1.0 (−2.2; 0.2)|
| BMI (kg/m2)||33.8 (33.2; 34.4)||31.9 (31.4; 32.4)||1.9 (1.1; 2.7)|
| Obesity (%)||All||65.9 (59.5; 72.2)||34.1 (27.8; 40.5)|
Table 3 shows the linear regression equations that we used to calculate individuals’ BMI from reported data. Mean BMI was calculated correctly by use of equation 1 for men and women. Among men and women, the prevalence of obesity was 1.6% and 1.0% lower when based on an equation with reported BMI, compared with the obesity prevalence rate calculated from measured data (i.e., 13.8% and 9.0% of the true obesity prevalence rate). When educational level, age, reporting weight with a digit preference, and smoking status were added to the equations, obesity prevalence rates were still underestimated by 12.9% and 8.1% among men and women, respectively (Table 3).
Table 3. Mean BMI and prevalence of obesity as calculated from equations that were calculated from the present data applied at individually reported data
| ||Mean BMI (kg/m2)||Obesity (%)|
|Men (n = 1749), calculated by use of measured values*||25.8||11.6|
| Equation 1-A: −0.1120 + 1.0212 × reported BMI||25.8||10.0|
| Equation 1-B: −0.6471 + 1.0155 × reported BMI + 0.0488 × education + 0.0100 × age + 0.0428 × rounding + 0.0288 × smoking||25.8||10.1|
| || || |
|Women (n = 1835), calculated by use of measured values*||24.9||11.1|
| Equation 2-A: −0.7041 + 1.0561 × reported BMI||24.9||10.1|
| Equation 2-B: −0.7253 + 1.0469 × reported BMI − 0.0358 × education + 0.0043 × age +0.2940 × rounding + 0.0397 × smoking||24.9||10.2|
The variance among studies in difference between measured and reported obesity prevalence rates is large, varying from 0.0% to 49.6% as percentage of the true obesity prevalence rate (Figure 1). In one study, where underestimation of obesity was nearly absent, the questionnaire to report body weight was sent out 2 weeks before the clinic appointment date (22). Any consistency between the obesity prevalence estimation based on reported data and the size of underestimation could not be detected.
Figure 1. Underestimation of obesity prevalence rates as a percentage of the true obesity prevalence rate in various studies. Larger data points represent larger studies. Numbers denote references. ps, present study.
Download figure to PowerPoint
Table 4 shows equations to calculate individual levels of BMI from reported data that were presented earlier by other authors. Applying these equations to our individual levels of reported data led to different calculations of the mean BMI, varying from 24.7 to 26.2 kg/m2, and to different calculations of the obesity prevalence, varying from 4.5% to 12.4% (i.e., a 7% overestimation to a 61% underestimation as a percentage of the true obesity prevalence). Among women, calculations of the mean BMI varied from to 22.5 to 25.2 kg/m2, and calculations of the obesity prevalence among women varied from 2.9% to 11.4% (i.e., a 3% overestimation to a 74% underestimation as a percentage of the true obesity prevalence) (Table 4). Characteristics of studies presented in Figure 1 and Table 4 are listed in Table 5.
Table 4. Mean BMI and prevalence of obesity as calculated from equations published in earlier studies applied at individually reported data from the present study
|Reference||Equation||Mean BMI (kg/m2)||Obesity (%)|
|Men (n = 1749), based on measured data:*|| ||25.8||11.6|
| (9)||Height (m) = (7.1987 + 0.8865 self-reported height (inches) + 0.022 × age − 0.0004 × age) × 0.0254||25.8||10.4|
| ||Weight (kg) = (−4.1259 + 1.0185 × SRW (pounds) × 0.4536|| || |
| (11)||Height = 15.302 + 0.923 × self-reported height − 0.052 × age||26.2||12.4|
| ||Weight = 0.561 + 1.012 × SRW + 0.006 × age|| || |
| (11)||BMI = 0.145 + 0.996 × self-reported BMI + 0.017 × age||26.2||11.5|
| (12)†||Height = 16.90 + 0.907 × self-reported height||24.8||5.0|
| ||Weight = 8.85 + 0.872 × SRW|| || |
| (12)†||BMI = 3.30 + 0.848 × self-reported BMI||24.8||4.5|
| (16)||Height (m) = [4.89 + 0.93 × self-reported height (inches)] × 0.0254||25.7||10.1|
| ||Weight (kg) = [−3.79 + 1.03 × SRW (pounds)] × 0.4536|| || |
| (19)||Height = 19.90 + 0.90 × self-reported height||24.9||5.8|
| ||Weight = 3.96 + 0.95 × SRW|| || |
| ||BMI = 2.60 + 0.87 × self-reported BMI||24.7||4.5|
| (22)||Height = 22.120 + 0.880 × self-reported height − 0.0435 × age + 0.0217 × SRW||25.4||7.7|
| ||Weight = 2.485 + 0.979 × SRW − 0.7086 (current smoker) + 0.0993 (ex-smoker) + 1.4034 (diabetic)|| || |
| (13)†||Height = 16.7 + 0.89 × self-reported height + 0.037 × SRW||25.2||6.1|
| ||Weight = 4.2 + 0.94 × SRW|| || |
| (13)†||BMI = 2.292 + 0.893 × self-reported BMI||25.0||5.3|
| || || || |
|Women (n = 1835), based on measured data:*|| ||24.9||11.1|
| (9)||Height (m) = [7.4583 + 0.8745 × self-reported height (inches) + 0.0424 × age − 0.0007 × age] × 0.0254||25.0||10.4|
| ||Weight (kg) = [−3.1974 + 1.0438 × SRW (pounds) − 0.0175 × age] × 0.4536|| || |
| (11)||Height = 27.096 + 0.853 × self-reported height − 0.069 × age||24.9||9.4|
| ||Weight = 0.444 + 1.010 × SRW + 0.006 × age|| || |
| (11)||BMI = −0.631 + 1.008 × self-reported BMI + 0.022 × age||24.8||9.5|
| (12)‡||Height = 22.92 + 0.886 × self-reported height||22.5||2.9|
| ||Weight = 6.72 + 0.87 × SRW|| || |
| (12)‡||BMI = 3.21 + 0.830 × self-reported BMI||23.4||4.0|
| (16)||Height (m) = [4.10 + 0.94 × self-reported height (inches)] × 0.0254||25.2||11.4|
| ||Weight (kg) = [−8.80 + 1.10 × SRW (pounds)] × 0.4536|| || |
| (19)||Height = 37.97 + 0.77 × self-reported height||24.0||5.4|
| ||Weight = 7.46 + 0.87 × SRW|| || |
| ||BMI = 5.57 + 0.74 × self-reported BMI||23.5||3.1|
| (22)||Height = 18.684 + 0.900 × self-reported height − 0.0422 × age + 0.0101 × SRW||24.4||7.6|
| ||Weight = 0.8759 + 1.0006 × SRW|| || |
| (13)†‡||Height = self-reported height + 0.033 × age||23.3||5.1|
| ||Weight = 0.95 × SRW + 0.041 × age|| || |
| (13)†‡||BMI = 1.835 + 0.893 × self-reported BMI||23.5||5.1|
Table 5. Characteristics of studies presented in Table 4 and Figure 1
| || || || || || ||Obesity prevalence|
| || || || || ||Men||Women|
|First author||Number of men||Number of women||Age||Period between health interview and examination||Measured||Reported||Measured||Reported|
|Bolton-Smith (22)||765||860||25 to 64||<2 Weeks||20.5||20.0||19.0||19.0|
|Boström (6)||1440||1768||18 to 84||4 to 6 Months||6.3||4.1||12.1||7.1|
|Flood (14)||94||133||16 to 85||0 to 6 Months||15.0||11.0||21.0||14.0|
|Hill (4)||1007||1251||16 to 64||1 to 4 Months||13.5||6.8||12.9||7.8|
|Kuskowska-Wolk (13)*†||119||182||16 to 84||Same day||14.3||10.1||13.2||11.5|
|Kuskowska-Wolk (12)*||1890||1500||18 to 84||4 to 6 Months||6.6||N.p.||12.5||N.p.|
|Niedhammer (15)||5342||1845||35 to 50||<6 Months||10.8||8.6||7.0||5.7|
|Nieto-García (28)‡||3507||3948||20 to 79||Same day||14.6||11.6||14.6||11.6|
|Pirie (16)||1608||1799||20 to 59||<1 Month||N.p.||N.p.||N.p.||N.p.|
|Roberts (10)||806||816||18 to 64||3 to 8 Months||8.1||6.3||9.8||7.6|
|Rowland (9)§||5396||5888||20 to 74||Few weeks||8.2||N.p.||11.6||N.p.|
|Spencer (11)||1870||2938||35 to 76||Few weeks||14.9||9.7||14.7||11.3|
|Stewart (8)‡||955||518||35 to 65||N.p.||9.3||6.2||9.3||6.2|
|Visscher (ps)||1809||1882||20 to 59||<3 Months||11.5||8.5||11.0||7.7|
- Top of page
- Research Methods and Procedures
The true prevalence of obesity was underestimated more severely than the true mean BMI, when reported body weight and height were used, partly because underreporting body weight occurred mostly in the obese. Although individual rates of underreporting were small, on average, obesity was underestimated by 26.1% in men and by 30.0% in women as a percentage of the true obesity prevalence. Linear regression equations that we calculated from the current study led to accurate estimations of the mean BMI, but the obesity prevalence was still underestimated when using these equations, by 12.9% and 8.1% of the true obesity prevalence rate among men and women, respectively.
Comparing published studies confirmed that reported body weight and height cannot substitute for measured body weight and height when estimating obesity prevalence for the following reasons: 1) Variation in underestimation of obesity prevalence by use of reported data is large and inconsistent. Thus, it appears to be unfeasible to produce a correction factor that allows the estimation of true obesity prevalence from a reported prevalence rate. 2) It is also inappropriate to use equations to calculate individuals’ BMI from their reported data. Equations may be appropriate for the estimation of mean body weight and in some studies for the estimation of the obesity prevalence rate in the same populations from which equations were calculated, but such equations should not be used across populations. It would be coincidental if the population of one's interest would be similar to the population of one of the few studies presenting equations. The most important barrier in making an appropriate choice for a suggested equation is that the strongest determinant of underreporting is unknown, namely true body weight.
Measured body weight was the most important determinant of underreporting body weight. High age and reporting weight with a digit preference in women were also slightly associated with misreporting, but adding these issues into the equation for calculating “true” BMI together with educational level and smoking did not improve our estimations of the obesity prevalence rate. Subjects older than 60 years tend to overreport their body height when their body height has declined since their last measurement (5, 9, 13). A relatively high level of education is a potentially more important determinant for underreporting body weight in women than in men because thinness may be more desirable in highly educated women (3, 15). A digit preference when reporting body weight was slightly more common in obese than in non-obese subjects. Obese women rounded their weights to the lower 5 kg rather than to the higher 5 kg. A history of dieting and degree of restrained eating has been reported as determinants of underreporting (3, 20, 29).
Frequent weighing, at least once per month, is reported to lead to more accurate reporting of body weight (14). Although we do not conclude that self-reported body weight and height may be valid alternatives to measuring body weight and height, it seems advisable to ask subjects to weigh themselves before they report their body weight and height in an interview or questionnaire. Spencer et al. (11) hypothesized that an alternative may be to measure at least a few subjects per quantile of the BMI distribution.
Use of self-reported data is of concern in large monitoring studies that are often meant to be nationally representative. For instance, from two representative studies from the United States, it has been reported that the obesity prevalence rate was 20.9% when based on reported data from a telephone survey (30) and 30.5% when based on measured data from a health examination (31). It is also possible that selective participation affected these differences. Consequently, population-based fractions of obesity-related consequences will be underestimated when based on obesity prevalence rates that are based on reported data. With an assumed relative risk of obesity for coronary heart disease of 2.5 (32), the fraction of coronary heart disease attributable to obesity would be 24% when based on reported data and 31% when based on measured data. Besides general underestimations of the obesity prevalence, Boström et al. (6) noted that wrong conclusions could be drawn regarding socioeconomic differences in obesity when differences are studied on the basis of self-reported data. If underestimation of body weight is higher in those with a high educational level, the social gradient in obesity prevalence rates will be overestimated with the use of reported data. Also of methodological concern is the use of reported body weight in epidemiological studies linking body weight to health outcomes. Specific underreporting in the obese may lead to an attenuation of relationships between obesity and health outcome measures. The relation between obesity and asthma, for instance, was attenuated when BMI was reported rather than measured (33). Odds ratios of obesity for asthma were 2.5 (95% CI, 1.1 to 5.9) and 2.3 (95% CI, 1.5 to 3.8) among men and women, respectively, when obesity status was defined on the basis of measured body weight and height and were only 1.7 (95% CI, 1.1 to 2.7) and 1.3 (95% CI, 0.6 to 2.9) when obesity was defined on the basis of reported body weight and height (33). Furthermore, use of reported body weight and height was inappropriate for estimating individuals’ obesity status. For obese men and women, based on measured body weight and height, body weight was underreported by 3.9 and 4.2 kg, respectively. Thus, 34.1% of the obese men and women would not have been identified as obese when based on self-reported body weight and height. It has been argued that using self-reported data as inclusion criteria for obese subjects in, for instance, weight loss studies may lead to selection bias of study participants (18, 34). Sensitivity to detect obesity may decrease with older age (28).
Although reported data do lead to underestimations of the obesity prevalence, reported data may lead to smaller biases when estimations of increases in obesity are studied on a yearly basis. In The Netherlands, the time trends in obesity have been similar when based on measured data (35) and when based on reported data (36). Flood et al. (14) concluded that periodic sub-studies of the validity of self-reported data are needed to indicate the extent to which the bias of self-reported data is changing over time.
Time delay between reporting and measuring body weight could be 4 to 6 months in referenced studies (4, 6, 10, 12, 14, 15), and it has been argued that body weight and height could change dramatically in such a period, especially in younger subjects (37). We propose that body height will not change in adults in such short periods, and mean changes in weight are usually small and could not explain large values of underreporting obesity. The studies that are compared in the present study were all performed in adults. Age ranges and educational status were similar, but not identical, in the various studies. It should be noted that age and educational status were not important determinants of underreporting body weight. It is more relevant to monitor obesity prevalence than the mean values of BMI, as the prevalence of obesity is increasing more rapidly over time than the mean BMI (35, 38).
- Top of page
- Research Methods and Procedures
Risk Factors and Health in The Netherlands, a survey on Municipal Health Services, was financially supported by the Ministry of Public Health, Welfare, and Sports of The Netherlands and by the Ministry of Economic Affairs. This work was supported by The Netherlands Heart Foundation within the framework of The Netherlands Research Program Weight Gain Prevention (Research Grant 2000Z002 to T.L.S.V.). We thank the contact persons and the fieldworkers of the Municipal Health Services in The Netherlands for contribution to the data collection for the study. The Project Committee consisted of J. van den Berg, F. Otten, and D. Hoezen from Statistics Netherlands (CBS), Division Social and Spatial Statistics, The Netherlands, J. Seidell, Ir. L. Viet from the Institute of Public Health and Environment (RIVM), and T. Coenen and Ir. H. van Veldhuizen from The Netherlands Association for Community Health Services. Data management was performed by J. Smolenaars and F. Frenken (CBS) and A. van Kessel and L. Viet (RIVM). Secretarial assistance was provided by Th. van den Brink, and laboratory assistance was performed by M. van Hemert, B. Elvers, and A. Wouters. We thank the people of the Centre of Infectious Diseases of the RIVM and the people from the Statistical Analysis of the CBS for their assistance. Above all, we thank all of the participants who participated in the study. We thank Elizabeth Spencer for sharing information regarding her research paper.