In the present study, factor structure and construct validity of a brief measure of eating behaviour, the Three Factor Eating Questionnaire Revised -18 (TFEQ-R18) , was examined in a population-based sample of Finnish adolescent and young adult females . Originally, based on psychometric analyses using data from Swedish obese men and women, Karlsson and colleagues shortened and revised the 51-item-Three Factor Eating Questionnaire , into a version with 18 items . Our analyses of TFEQ-R18 -data from the 2 997 young Finnish females with varying body weights produced a factor structure that corresponded to the one found by Karlsson et al : six items loaded high on the factor "Cognitive Restraint", nine items loaded high on "Uncontrolled Eating", and three items loaded high on "Emotional eating". Construct validity of the TFEQ-R18 was good.
Our findings corroborate earlier results suggesting that the TFEQ-R18 is a valid measure of eating behaviour not only in the obese but also in the general population. In samples of French adolescents and adults, multitrait/multi-item scaling analyses showed satisfactory internal consistency of the French translation of TFEQ-R18 . The French study reported very similar internal consistency reliability coefficients (Cronbach's alphas) for the three scales to the ones reported by the designers of the TFEQ-R18 .
TFEQ-R18 scores and body weight
The secondary aim of our study was to analyse connections between body weight and TFEQ-R18 scores. Connections between eating behaviour and body weight have been extensively studied, especially in the field of obesity research. However, to the best of our knowledge, there are but two earlier studies that explored this connection in samples with varying weights, and in which TFEQ-R18 was used as a measure of eating behaviour. Elfhag and Linné studied eating behaviour and relative weight of Swedish women, aged 35 to 65 years, and their adolescent children, aged 15 to 18 years . De Lauzon-Guillain and colleagues analysed TFEQ-R18 responses and several measures of adiposity in a sample of French adults and adolescents, over a two-year period . In the latter report, the main focus was on cognitive restraint.
We found that of the three factors of the TFEQ-R18 questionnaire, cognitive restraint and emotional eating were connected with body weight in girls and young women. Higher scores of cognitive restraint and higher scores of emotional eating were associated with a higher BMI. These results are similar to the cross-sectional findings of Elfhag and Linné : both cognitive restraint and emotional eating were positively correlated with BMI in adolescent girls, in adolescent boys, and in their mothers.
Of the three factors of TFEQ-R18, we found no connection between uncontrolled eating and body weight, when BMI was analysed as a categorised variable. When analysing BMI as a continuous variable, there was a statistically significant (p < 0.001), but a very weak (r = 0.063) positive correlation between uncontrolled eating score and BMI. In the latter analysis, the correlation probably reached significance due to the large sample size. In the Swedish adolescents, uncontrolled eating had no connection with BMI, whereas in the adult women, there was a positive correlation between BMI and uncontrolled eating . Our sample was on average slightly older than the girls, but clearly younger than the women of the Swedish study. Perhaps the connection between higher body weight and higher level of uncontrolled eating in females starts to appear with increasing age.
These cross-sectional findings raise the question of whether it is the eating behaviour that predicts body weight, or vice versa. Do the tendencies of cognitive restraint and emotional eating make people gain weight. Or does a higher body weight make people eat differently? The two-year-follow-up by de Lauzon-Guillain and colleagues  suggest that the latter might be true, at least in case of cognitive restraint. In the French sample, initial scores of cognitive restraint were not connected with subsequent adiposity changes, neither in adults, nor in adolescents. However, in all age groups studied, higher values of initial adiposity – BMI, waist circumference, sum of skinfold thicknesses, and percentage body fat – predicted a larger increase in cognitive restraint score over the two-year period . Thus, differences in adiposity level seem to precede differences in cognitive restraint.
Since the present study was carried out among participants of a phase III human papillomavirus (HPV) vaccination trial, which aims at following the efficacy of a vaccine against the HPV virus and the later development of cervical carcinoma, representativeness of our sample is limited even if the invitation was population-based. The subjects had volunteered to participate in the vaccination trial as well as in our study on eating behaviour. Thus, these young females are a selected group, likely to be more health conscious than their peers among the general population. On the other hand, an ongoing questionnaire study with 4 438 respondents found that the young women who participated in the HPV vaccination trial did not differ significantly from young women who did not participate, regarding certain measures of living conditions and quality of life (e.g. physical functioning and well being, emotional wellbeing, social functioning, life habits and sexual health) (Woodhall et al., unpublished data). Therefore, all in all, the phase III HPV vaccination trial provided a unique opportunity for us to attain a large and representative sample of young females from different parts of Finland. For example, well-organised weight and height measurements of the nearly three thousand subjects by professional nurses – instead of having to settle with self-reported weight data – would hardly have been possible without the organisation of the HPV vaccination trial.
We also slightly modified one of the items of the original TFEQ-R18 questionnaire. The question "When I smell a sizzling steak or a juicy piece of meat, I find it very difficult to keep from eating, even if I have just finished a meal." (item 1) was replaced with "When I smell a delicious food, I find it very difficult to keep from eating, even if I have just finished a meal." The item is supposed to measure the tendency to uncontrolled eating in the absence of hunger, when tempted by external stimuli. In the Finnish culture, steaks and meat are not necessarily considered the most desired foods – definitely not among girls and young women. Steaks are thus a poor example of a tempting food in this population, and the original item needed to be changed in order for it to produce valid responses. The data we gathered using the TFEQ-R18 behaved in the analyses in a very similar manner when compared to earlier analyses of TFEQ-R18 data [6, 8, 18], suggesting that the instrument was valid despite the slight modification of one of the eighteen items.