Sample and variables

We used the database from the Cross National Student Health Survey (CNSHS), consisting of 5,900 records of university students from seven countries in Europe [16]. The participating universities were: the University of Bielefeld, Germany (DE), the University of Lublin, Poland (PL), the University of Sofia, Bulgaria (BG), the Navarra Public University and the University of Navarra, Spain (ES), the Universities of Kaunas, Lithuania (LT), the University of Southern Denmark (DK) and Hacettepe University, Turkey (TR). For ease we use country names in the text and country codes in the tables. Data from four countries were collected between 2003 and 2005, while data from Spain and Lithuania were collected between 1998 and 2000. In Spain, students were recruited through an advertisement at the university campus, and about 28% of all first-year students at that university participated in this study. In all other countries, first-year students from randomly selected courses were invited to complete a self-administered questionnaire during the last 10 minutes of their lectures. Response rates in these countries ranged from 85% to 97%. Participation was voluntary and anonymous. Permission to conduct the study was obtained from the participating institutions. Students were informed that by filling out the questionnaire they were providing informed consent for the participation in the study. They were also instructed that they can withdraw from the study at any point by not returning the questionnaires. In Spain, where additional biomedical measurements were conducted and blood samples were obtained, the study was approved by the local ethics committee, and written informed consent was obtained from all participants.

Among the 5,900 participants, 558 (9%) did not report either their weight or height. The fraction of missing responses was substantially higher in Spain (32%) than in other countries (2-8%). This is probably resulting from the fact that students in Spain completed the questionnaire before having physical exams including weight and height measurement and either did not know their exact weight and/or height or did not feel that it was necessary to report the information. Students with missing information on weight or height did not differ with respect to measured weight from students who self-reported weight and height. Because weight and height were only measured in Spain, this analysis is based on self-reports and restricted to students who reported weight and height. In the Spanish sample, students on average underreported weight by 0.60 kg and overreported height by 1.07 cm (unpublished data). This resulted in an average underreporting of BMI of -0.49 kg/m².

In order to assess students' perceived weight, they were asked: "Do you consider yourself much too thin, a little too thin, just right, a little too fat or much too fat?" BMI was calculated from self-reported weight (in kilograms) and height (in centimetres).

Statistical analysis

Because we relied on students' self-reported height and weight, we estimated the extent of guessing or rounding by examining the data for last digit preference, e.g., overreporting of numbers ending in zero or five [17]. We calculated the proportion of responses with last digit "0" or "5" by country and by gender. To assess whether the proportions differed between countries within each gender we used DerSimonian Laird estimator [18]. DerSimonian Laird estimator is a test for heterogeneity across different samples commonly used in meta-analyses; it is equivalent to testing whether the variance of random effects is different from zero. The test was conducted using software META^® [19]. Additionally, we studied the differences in last digit preferences across BMI categories in male and female students. Self-reported weight and height and calculated BMI were displayed as means for each country and gender. Given the sample sizes in different countries (stratified by gender), differences of ≥ 3 kg in weight, ≥ 2 cm in height and ≥ 1 kg/m² in BMI are statistically significant at p < 0.05 (statistical significance of pair-wise comparisons is not reported in the text). These differences also appear to be reasonably meaningful. In order to provide further information, we also estimated the range of effect sizes in terms of Cohen's d for differences between genders and across countries for weight, height and BMI.

According to WHO guidelines [20], students were classified as underweight (BMI < 18.5 kg/m²), normal weight (BMI of 18.5-24.9 kg/m²), overweight (BMI of 25.0-29.9 kg/m²), or obese (BMI ≥ 30.0 kg/m²), and the proportions of students in these categories were computed in each country separately for males and females. Given that BMI based on self-reported data is usually biased downwards [21], we conducted a sensitivity analysis increasing the reported weight by 1 kg in all participants, which is somewhat larger than the underreporting observed in Spain. This increase in weight resulted in an increase in BMI by 0.30 to 0.37 kg/m², which is about twice as much as the underreporting bias seen in our own validation study conducted in Germany [22], and somewhat less than the estimate of underreporting in the Spanish students above.

Finally, we investigated how perceived body weight was related to the BMI reported by students. Three separate models with dichotomous responses were employed: 1) "just right" versus the remaining categories, 2) "much too thin" or "little too thin" versus the remaining categories, 3) "much too fat" and "little too fat" versus the remaining categories. The probability of a given response across the BMI spectrum was modelled using non-parametric regression with locally weighted polynomial fit (loess function, implemented in R library gam [23]. Non-parametric regression allows the assessment of the form of the association between the independent and dependent variables. The analysis was conducted separately for both genders and each country; the existence of statistical differences across these strata was tested using an interaction term in a joint model. Again, the analysis was repeated with BMI calculated from reported weight increased by 1 kg for all participants.

Last digit preference for self-reported weight or height

There was a considerable preference for last digits 0 and 5 in self-reported weight or height, as in all cases the fraction of such reports considerably exceeded the expected 20% (Figure 1). There was also some variation between the countries and by gender in the extent of this preference (heterogeneity test p < 0.001), but in general, the extent of preference was rather similar. Female students had slightly lower last digit preference for weight than male students. The extent of last digit preference across both genders was lowest in Lithuania and highest in Turkey. The lowest levels of last digit preference were in height among male students in Germany and Spain. Additionally, we investigated whether the extent of last digit preference depended on BMI category based on self-reported weight and height. Since there was little variation across countries, we performed a pooled analysis across all countries. A clear difference appeared only in females and only with respect to weight: overweight femalgs were more likely to display a last figit preference than normal or underwekgjt females (Figure 2). Inspection of the last digit of eivher reported weight or height suggest that occurrence of rounding was most likely for digits "1" and "9" rounded to "0" and for digits "4" and "6" rounded to "5".

Differences in height, weight and prevalence of overweight between countries and genders

Country differences in height and weight were slightly more pronounced in female than in male students (partial eta-square among females 0.11 for height and 0.10 for weight, among males 0.06 for height and 0.09 for weight). Female students were shorter in Southern (Spain, Turkey) and Eastern European (Poland, Bulgaria) countries than in Western or Northern European countries (Germany, Denmark, and Lithuania) (see Table 1). In terms of Cohen's effect sizes, the pairwise differences within the above groups (Southern, Eastern etc.) were in most cases below d = 0.2, which means small effects; in contrast differences between countries from discordant groups were d > 0.2. Lithuanian women were particularly tall, with a difference of 6 cm in height to either Spanish or Turkish female students (effect size of 0.85). The pattern was similar for the height of male students, with the exception of Lithuania. Since males from Lithuania tended to be shorter, the difference between both genders was exceptionally small in Lithuania.

There were also some differences with respect to weight: female and male students had the highest mean weight in Germany and Denmark. In the case of weight, the above defined grouping of countries was less demarked, with somewhat larger differences within groups (difference in mean weight between females from Turkey and Spain of 3 kg; d = 0.4), but still substantial differences across the spectrum of all countries: maximum difference of 8 kg across countries, d ≥ 0.7. Difference in weight by gender was smallest in Lithuania (d = 0.9) and largest in Bulgaria (d = 1.8). Evaluation of BMI based on self-reported weight and height showed that the differences between countries were somewhat smaller for male students (maximum difference of 1.7 kg/m², d = 0.63) and somewhat larger for female students (maximum difference of 2.5 kg/m², d = 0.78) with an exceptionally low BMI in female Bulgarian students.

Differences in perceived weight

Although the findings indicated a high proportion of normal weight among the sample, only 32% to 68% of students considered their weight "just right" (Table 2). Between 22% (Turkey) and 51% (Germany) of female students considered themselves "a little too fat". Only in Turkey was there a substantial fraction of female students who considered themselves too thin. Among males, between 11% (Lithuania) and 38% (Germany) of the sample considered themselves "a little too fat", with substantial proportions in all countries who considered themselves "a little to thin" (13% in Denmark, 30% in Poland).

While Table 2 simply compared the distributions of BMI and perceived weight status by gender and country, the discrepancies between BMI and perceived weight status and the gender differences become even more apparent when they are considered jointly. In Figure 3, we combined students from Denmark, Poland, Bulgaria, Turkey and Spain because their country plots were very similar. We depicted Germany and Lithuania separately because they deviated from the general pattern.

In the first panel of Figure 3, less than 70% of students considered their weight "just right" for any given BMI. More significantly, female students felt "just right" at a lower BMI than their male peers did. Around 20% of females with BMI of 20 kg/m² considered themselves "too fat", and the percentages increased to 60% for a BMI of 22.5 kg/m². Male students rarely felt "too fat" below BMI of 22.5 kg/m². Conversely, most male students felt too thin with a BMI of 20 kg/m².

In the German sample, only about 40% of male students considered their weight "just right", a much smaller proportion than in all other countries. In Lithuania, the proportion of women considering their weight "just right" remained reasonably high for normal BMI. However, this was also accompanied in Lithuania by a relatively high proportion of female students with BMI above 25 kg/m still considering their weight "just right".

Strengths and Limitations

As strength, this study included a relatively large number of students from several European countries. With the exception of Spain, students were not explicitly informed that the survey included questions about weight. Therefore, it is unlikely that students declined participation because of this aspect. Still, there may be a selection bias, because students not interested in the health survey may differ in respect to their weight and their weight perceptions from their peers who participated in the study. Due to the high response rates achieved, this bias seems to be limited in all countries except Spain, where the response rate was substantially lower. However, findings from Spain were well in line with other countries, suggesting that the selection bias related to self-reported weight and weight perception may be similar in all countries included in this analysis.

As in the International Health and Behaviour Survey and the Health Behaviour of School Aged Children studies, our measure of BMI was based on self-reported weight and height. The reliability and validity of self-reported weight and height in different populations has been controversially discussed [34–41]. Most studies found a limited underestimation of BMI but with partly substantial effects on prevalence estimates for BMI categories. A recent review of studies conducted among adolescents in the United States indicated mean differences in self-reported versus directly measured BMI of -2.3 to 0.2 in females and -3.0 to -0.1 in males. While differences between self-reported and directly measured BMI were very small in nationally representative surveys, they were substantially larger in convenience samples or locally-based surveys [21]. Among other variables, the differences between reported and measured BMI might depend on education and age. An underestimation of 0.98 kg/m² in females and 0.75 in males for BMI was found in Greek adolescents [38]. A recent analysis using German HBSC sample demonstrated that adolescents, especially younger adolescents often do not report their weight or height, possibly because they do not know them [42]. Large proportions of missing BMI data were found in HBSC samples from many countries as well [14]. University students could be considered as the population group with more accurate, maybe best possible reporting, given their high education status and age range in early adulthood. In our own validation study in a German sample of students, we found an acceptable agreement between self-reported and measured BMI (mean difference of 0.18 kg/m²) [22]. Depending on the distribution of BMI, in some countries the underestimation might affect the prevalence of some BMI categories more than in others, but as demonstrated in the sensitivity analysis, correcting for a homogenous underestimation of weight did not change the qualitative differences observed across countries. More complex assumptions about underreporting of weight and possibly overreporting of height could result in further changes, therefore our estimates of the prevalence of BMI categories should be treated with caution. There could be cultural differences in the accuracy of reported BMI, but our analysis on last digit preferences for self-reported height and weight does not support this notion. On the other hand, the underreporting would affect findings presented in Figure 3 only slightly, lending further support to studies that use self-reported BMI for correlational analyses (in contrast to the estimation of prevalence of overweight) [43, 44].

As many other studies [4, 8] have done, we assessed perceived weight using a single question. Future research should use more sophisticated instruments to assess body perceptions with regard to muscularity, height and body fat distribution [33].

Our data were generated from student samples at one or two universities per country and may not be representative of all students of the respective countries. A comparison with representative data on students' age and gender distribution [45] confirms that Danish and German students are on average older than students from the other countries that are included in this analysis, and that there are more female than male students. Nevertheless, any distortion of the gender distribution is not likely to affect our results as the analysis was stratified by gender. Finally, the study was not conducted in all countries in the same year and while we do not expect large changes during a five-year period, some changes might have occurred.