Physical activity and ethnicity
In our study, boys were more physically active than girls, as has been reported earlier [6, 7, 37]. We found ethnic differences in activity level, an observation that is consistent with previous findings from the UK  and USA [13, 18], although other studies in the USA reported no differences  or higher physical activity level in some ethnic minority groups . The difference between the USA and Europe may reflect differences in the immigrants' ethnicity, history, and time since immigration. Discrimination and racism are other factors that might cause differences in sport participation between ethnic minority and host adolescents [38, 39]. Fear of being exposed to racism could be keeping ethnic minorities away from organized sport . However, our results do not allow us to conclude whether racism influences the choice of physical activity type.
As observed in other studies [13, 18], we found that ethnic differences in physical activity were more apparent in girls than in boys. Differences in physical activity may reflect the influence of factors such as the religion and culture in the country of origin. In our study, 96% of the ethnic minorities came from non-Western countries, the largest group being from the Indian subcontinent. The difference between boys and girls could relate to gender segregation in some religions (Islam), girls having more household responsibility, or stricter rules set by parents [41–43]. The ethnic minority girls' low activity level might also relate to the structure of organized sports in Norway, which includes fewer differences according to gender than in many of the ethnic minorities' countries of origin . Another contributing factor could be the low physical activity level among ethnic minority women in Oslo . Parents' physical activity patterns probably influence their children's physical activity through modelling, social influence, and social support .
Physical activity and socio-demographic factors at age 15 and 18 years
Consistent with other studies, [24–26] we found few significant associations between socio-demographic factors and physical activity level. Such associations seem to be dependent on the SES measure used and the characteristic of the subgroup being studied.
One relatively consistent finding in our study was the lower physical activity level among ethnic Norwegians who perceived family economy as poorer than other Norwegian families. Those who perceived themselves as having low family income might be restricted in their physical activity choices and opportunities because of the cost involved.
The lack of association between physical activity and SES in ethnic minorities may indicate a different relationship between SES and health and disease, and a different influence on health behaviour in ethnic minorities than in ethnic Norwegians. A Norwegian study focusing on determinates of diabetes in different ethnic adult groups reported a negative association between the prevalence of diabetes and SES among ethnic Norwegians and Westerners but almost no association in ethnic minorities .
Change in and stability of physical activity from age 15 to 18 years
We found a decline in mean hours per week of physical activity from age 15 to 18 years, which was similar in ethnic Norwegians and ethnic minorities. In the UK, Asian adolescents and black girls are less active than white girls at age 11–12 years, and this difference does not change over the next five years . McMurray et al.  observed a similar decline in physical activity from age 8 to 16 years in African-American and Caucasian girls but a greater decline in Caucasian boys than in African-American boys. Another longitudinal study from the USA  found a substantial decline in physical activity level that was higher in black girls than in white girls.
The relatively low κ scores in our study suggest low stability of physical activity levels within groups. Anderssen et al.  used a similar physical activity questionnaire with youths in the western part of Norway and reported κ values, based on tertiles, of 0.26 for boys and 0.21 for girls over a three-year period (age 16–19 years). The only study that assessed physical activity stability in different ethnic groups (Caucasians and African-Americans) reported low κ values (0.03–0.22) . The discrepancy in results might be because κ appears to be higher for shorter time periods, whereas our study and the study by Anderssen et al.  studied stability over a three-year period, and McMurray et al.  studied stability over seven years. We also dichotomized physical activity level at two times, whereas the other studies divided the physical activity variable into three groups.
McMurray et al.  also reported that more Caucasian than African-American youth remained in the low physical activity group during the follow-up. In our study, the highest percentage of persistently low physical activity was found among the ethnic minority girls. As suggested previously, the different findings might be explained by the different immigrant histories and cultures of origin of the ethnic minorities in the USA and in Norway. The discrepancy in results may also be caused by different sample sizes, age, and the definition of "low physical activity".
Socio-demographic factors and change in physical activity
The only socio-demographic factors that were associated with change in physical activity level were mother's education in girls, and father's income in boys. The few associations observed might reflect the influence of more important circumstances experienced by this group of youth. These adolescents had experienced changes in schools, increasing homework level, and the biological, social, and psychological changes that accompany puberty. These factors, alone or in combination, may affect physical activity significantly and might "overrule" the importance of socio-demographic factors.
Measuring physical activity level by questionnaire is associated with difficulties . The measure used to capture physical activity level in this study is general: "activities outside of school that make you feel sweaty and out of breath". Such a crude overall measure does not capture all the physical activities that promote health. However, simple, self-reported questions on overall physical activity have been used in several studies and correlated significantly with other activity measures [47, 48], and with other indicators of physical activity such as maximal oxygen uptake  and physical fitness . It seems reasonable to assume that the question captures the level of physical activity, although we do not know how accurate the adolescents were in reporting hours per week. However, we compared associations, change in, and stability of physical activity between groups and believe that the measure is reasonably reliable and valid for this purpose. Generally, the more unreliable the measure, the greater the chance of not finding differences or associations that exist. Therefore, we probably underestimated rather than overestimated any associations and difference between the groups.
Performing a large number of tests, as we did when studying the relationship between physical activity level and socio-demographic factors (aims 2 and 4), increases the risk of type 1 error. Hence, we emphasize the pattern of our findings and interpret the single significant associations with caution.
We also note the variety of ethnic minorities represented by our sample. Even though about 96% of the ethnic minorities in our study population were from non-Western countries, they were not a homogenous group. Studying all ethnic minorities together might conceal differences in physical activity levels between different ethnic groups.
The attrition is also a concern. Of those participating in the baseline study, 70% of the ethnic Norwegians and 54% of the ethnic minorities participated in the follow-up. Ethnic minorities participating at follow-up reported more physical activity at baseline than did the ethnic minorities lost to follow-up; there was no corresponding difference at baseline in the ethnic Norwegians. However, in the ethnic minorities, the physical activity level did not differ between those who completed the questionnaire after reminders compared with those who participated after the primary invitation, indicating that the selection probably did not influence the results substantially .