The study received ethical approval by research and ethics committees at the participating institutions. A representative sample of students was sought at all participating universities by selecting courses adequately representing the different departments/faculties, where self-administered questionnaires were distributed to all participants attending the selected classes/courses, and then collected after completion. Students were informed that participation was voluntary and anonymous, and that by completing the questionnaire, they agree to participate in the study. No incentives were provided for participation. Each questionnaire included a participant information sheet outlining the research aims and objectives. Students who remained in class to participate read the information sheet and, if they wished to participate, removed and kept it for future reference. In this manner it was ensured that the students who stayed in class to complete the survey consented to their participation and were not unwittingly coerced into the survey. Data were confidential and data protection was observed at all stages of the study. All data were computer-entered centrally at one site thus maximising the quality assurance while minimising potential data entry errors.

Data were collected during the academic year 2007–2008 at seven universities in England. The total sample comprised 3705 undergraduate students from England [University of Gloucestershire (n = 970, 43.6%_♂, mean age 23.3 ± 8.4 years), Bath Spa University (n = 485, 22.6%_♂, age 22.2 ± 6.9), Oxford Brookes University (n = 208, 10.8%_♂, age 31.6 ± 10.4), University of Chester (n = 993, 13.1%_♂, age 26.0 ± 9.2), Plymouth University (n = 169, 56.2%_♂, age 24.6 ± 7.2)]; Wales [University of Swansea (n = 406, 7.8%_♂, age 25.0 ± 7.4)]; and Northern Ireland [University of Ulster (n = 474, 8.2%_♂, age 25.2 ± 7.7)]. Based on the number of returned questionnaires, the response rates were ≈ 80%.

The study was a general student health and wellbeing survey similar to studies of student health implemented in several countries [73,74] and comprised 133 items with an average 15–20 minutes completion time. It included socio-demographic information (e.g., gender, age), self-reported health data, as well as questions of health behaviours (nutrition, PA, smoking and alcohol consumption), social support, and university study related questions.

Vigorous exercise (1 item): students were asked: “On how many of the past 7 days did you participate in vigorous exercise for at least 20 minutes?” Participants answered with a rating from 0 to 7 days. A cut-off of at least 3 days a week was used in the analysis based on guidelines of the American Heart Association (AHA) for vigorous PA [75].

Moderate exercise (1 item): “On how many of the past 7 days did you participate in moderate exercise for at least 30 minutes?” Participants answered with a rating that ranged from 0 to 7 days. A cut-off of at least 5 days a week was used in the analysis based on AHA guidelines for moderate PA [75].

Strengthening or toning muscles (1 item): “On how many of the past 7 days did you do exercises to strengthen or tone your muscles, such as push-ups, sit-ups, or weight lifting?” Participants answered with a rating that ranged from 0 to 7 days. The cut-off of at least 2 days a week was used in the analysis based on AHA recommendations [75].

Depressive symptoms: were measured using the Modified Beck Depression Inventory (M-BDI) [44–46]. The modification of the original BDI included two approaches: (a) the four items per symptom which assessed the specific symptom’s intensity in the original BDI were replaced by a single statement per symptom with a six point Likert scale measuring its frequency in the last 4 weeks (past few days in our questionnaire) (with the two extreme categories labelled as 0 = ‘Never’, 5 = ‘Almost Always’). Sample items include: “I feel sad”, “I feel I am being punished”, “I have thoughts of killing myself”, “I have lost interest in other people”, “I have to force myself to do anything”, “I am worried about my appearance”, “I have no appetite”, and, “I lost interest in sex”. The reduction in the number of items per symptom is consistent with another recent modification of BDI (BDI-II) [46]. The versions of the BDI compute a single score for individual respondents by summing their responses for all items of the scale. Through a German sample reflecting the general population and in selected subsamples [44,45], the authors of the M-BDI demonstrated its construct validity and measurement equivalence as compared to the original BDI. They also provided a cut-off score for screening for clinically relevant depressive symptoms at ≥35, corresponding to the 85th percentile of the representative sample of the German population [76].

Body image perception (1 item): assessed on a five-point Likert scale adapted from the Health Behavior in School-aged Children (HBSC) study [77] and recently used by others, e.g., El Ansari et al. [74] after a change of the word “fat” of the item to “overweight”, as it was felt that the latter term is less judgmental. Students were asked: “In your opinion are you…”, with five response options (“Far too thin”, “A little too thin”, “Just right”, “A little overweight”, “Very overweight”). For the analysis, the five options were re-coded into three binary variables (“Underweight”, “Just right”, “Overweight”).

Self-rated health status (1 item): “How would you rate your health in general?” with a 5-point scale response format (1 = “excellent”, 2 = “very good”, 3 = “good”, 4 = “fair”, 5 = “poor”) as used in the German Federal Health Survey [78] (similar wording used by American College Health Association [79].

Health awareness: (1 item): as used by others [80], an item asked students about their general awareness of their health: “To what extent do you keep an eye on your health?”, with a four-point response scale (1= ‘not at all’ to 4= ‘very much’)

Educational achievement (1 item): ): assessed on a five-point Likert scale adapted from the Health Behavior in School-aged Children (HBSC) study [77]. As our sample comprised university students, we modified the original item (how teacher rated the pupil’s performance) to a self-rating: “How do you rate your performance in comparison with your fellow students?” (five response categories, from 1 = ‘much worse’ to 5 = ‘much better’) [47].

We used SPSS 14.0 (SPSS Inc. Chicago, IL) to calculate frequencies and proportions and to conduct the statistical analyses. In order to present prevalences of students performing different levels of PA by university taking into account the varying male-to-female ratio of the samples at the different sites, we sex-adjusted the prevalences using direct standardization towards a male-to-female ratio of 30% to 70%. For several variables, some of the response options were combined to satisfy the assumption of adequate cell size for regression analysis. We performed multifactorial logistic regression analysis with enter mode in order to examine odds ratios for the association between depressive symptoms and different forms of PA (low, recommended moderate, recommended vigorous) as dependent variables after adjusting for potential confounding factors (gender, perceived health, health awareness, muscle strengthening, academic performance). While vigorous and moderate PA were defined as stated above, low PA was defined, in line with others, as “not participating in any vigorous or moderate PA or participating in only one day of moderate PA” [81]. We then stratified the analysis according to perceived body image (as ‘just right’ or as ‘overweight’) in order to explore whether perceived body image modified the association between depressive symptoms and different forms of PA. The ‘underweight’ category of perceived body image was not included in the stratified analysis due to inadequate sample size in this category. In order to determine the goodness-of-fit of each model to the survey data we used the Hosmer-Lemeshow test.

Table 1 shows that only 12.4% of our sample of students achieved the American College of Sport Medicine and AHA’s recommended guidelines [75] of moderate PA per week, while 33.1% achieved the recommended levels of vigorous PA. Roughly 20.6% had a depression score >80th percentile. Every second student perceived their own health as ‘very good’ or ‘excellent’, and the majority (84%) had health awareness and reported that they ‘kept an eye’ on their health. A small proportion (15.8%) rated their own academic performance at university as better than that of their peers. Slightly more than half the participants (59.6%) perceived their body image as ‘overweight’.

Table 2 shows the results of multifactorial logistic regression models for the factors associated with three different levels of PA as the dependent variables. Academic performance was not associated with any type of PA. Both forms of PA at recommended level were inversely related with the M-BDI score >5th quintile, but vigorous PA was additionally inversely related with M-BDI score at moderate level (>4th quintile). Students achieving PA guidelines, regardless of the type of PA, were significantly more likely to perceive their health as good (better than those who did not achieve the PA guidelines), to have higher health awareness/consciousness, to perform strengthening exercises, and to be males. In contrast, students with low PA level were more likely to be older, to have lower levels of perceived health and health awareness, and were less likely to perform strengthening exercise. In contrast to students being physically active at recommended level, being low active was not associated with depressive symptoms or with gender.

Table 3 shows the results of multifactorial logistic regression models for factors associated with three different levels of PA as dependent variables by body image perception (as ‘just right’ or ‘overweight’). The stratified analyses indicated that the association between depression and PA differed by body image:

In students perceiving themselves as just right, moderate (>4th percentile) and high (>5th percentile) M-BDI scores were inversely related with vigorous PA. However, in students who perceived themselves as overweight, the inverse association was only significant in those with high M-BDI scores.

Other factors associated with PA differed also by body image:

Academic performance was positively associated with low PA in students perceiving themselves as just right and negatively in students perceiving themselves as overweight

Roughly 12.4% and 33.1% of our sample of students achieved the recommended AHA guidelines [75] of moderate and vigorous PA levels of PA per week respectively. Both these values compared unfavourably with recent PA levels of college students from 4 universities in 3 Midwestern states in the USA, where 39% and 56% of the total respondents met the moderate and vigorous guidelines respectively (i.e., 56% of the total respondents participated in at least 20 minutes of vigorous physical activity on 3 or more days of the past 7 days, and participated in at least 30 minutes of moderate physical activity 5 or more days of the past 7 days) [81]. This discrepancy might be related to the samples’ characteristics. Seo et al.’s [81] participants were from a group of students who were enrolled in health-related courses, suggesting that those students might have had a pre-existing interest in health behaviours or that their current engagement in the course had influenced their activity levels. Although our sample had the majority of students from health related courses (e.g., nursing or health and sport related, there were other participants from other courses e.g., psychology and business courses). However, the levels recorded by our sample also seemed lower than the national estimates reported by researchers who used the 1995 National College Health Risk Behavior Survey, where 20% met the moderate guideline and 38% for met the vigorous guideline [82].

As regards strengthening exercises (≥twice per week), whilst 23.9% of our sample reported such exercises at least twice per week, such levels compared unfavourably with the 47.7% of students in the USA who reported that they exercised to strengthen or tone muscles at least 2 out of the past 7 days [83]. In connection with depressive symptomatology, roughly 20.6% of our sample exhibited a depression score >80th percentile. These levels compared favourably with university students in a range of western and eastern European countries (Poland, Bulgaria, Denmark and Germany) [84]. The proportion of students from those countries with MBDI scores ≥35 (corresponding to the 85th percentile) was above 40% (female students in Poland and Bulgaria), to around 25% (female students in Denmark and Germany); and for male students it fluctuated (22.8%-Germany, 12.1%-Denmark, 27.3%-Poland and 33.8%-Bulgaria) [84].

In connection to our research aim to assess the association between depression and PA, we expected to find a significant inverse association even when controlling for other factors. The results of our analysis showed that students performing PA at recommended level were less likely to have an M-BDI score >80th percentile, a limit that is an accepted indicator for depression, independent of potentially confounding factors. Therefore our findings support a protective effect of PA, or that physical inactivity may be a risk factor for depressive symptoms, which is in line with other cross-sectional [17–19] and prospective studies [20–22]. Although some longitudinal studies did not clearly demonstrate protective effects of PA programmes (see introduction) this can be result of ineffective programmes or weak implementation and does not necessarily contradict our findings. It is important to note however, that from our cross-sectional study we cannot conclude any cause-effect relationship: PA may lower depression or depression may lower PA.

Regarding the research aim to study dose-response relationship between PA and depression, we hypothesized that university students who achieve recommended guidelines of vigorous PA would exhibit lower levels of depressive symptomatology than students who achieve recommended levels of moderate PA. The highest level of depression would be expected in students not engaged in any level of PA. Our findings are supporting these hypotheses. While low PA level was not associated with depression, the protective effect of PA increased with PA intensity. Both forms of PA (moderate and vigorous) were inversely related with the M-BDI score above the 5th quintile, but vigorous PA was additionally inversely related with M-BDI score at moderate level (>4th quintile). These results suggested that the association between PA and depression may increase with PA intensity, which has also been found in some few other studies [26,59,60].

A sub aim of the analysis was to study the role of body image perception as a potential effect modifier of the relationship between PA and depressive symptomatology. We hypothesized that the inverse association between depression and PA might be stronger in students who perceive their body image as ‘just right’ when compared to students who perceive their body image as ‘overweight’. In line with this hypothesis we found that in students perceiving themselves as just right, moderate (>4th quintile) and high (>5th quintile) M-BDI scores were inversely related with vigorous PA. However, in students who perceived themselves as overweight, the inverse association was only significant in those with high M-BDI scores. This might be explained in the way that a ‘distorted’ body image would down modulate the inverse association between depression and PA. In other words, the positive effect of PA on depression would be down modulated by the negative impact of a ‘distorted’ body image on depression. In this sense a ‘distorted’ body image might be remotely compared to physical disability which represents a potentially powerful confounding variable as it is associated with both a higher risk of depression and lower levels of physical activity [19]. A practical implication of this result could be that PA programmes targeting persons with depressive symptomatology should include a component to improve body perception.

A second interesting finding regarding body image as effect modifier is that low PA was positively associated with academic performance in students perceiving themselves as just right, but negatively in students perceiving themselves as overweight. This suggested that students performing well were more likely to be sedentary when they were satisfied with their weight/body image, but more likely to be physically active when they perceived themselves as overweight.

This study has limitations and generalization of the findings requires caution. As a cross-sectional survey, the findings are associations not causations, with difficulty in determining the direction of the effects. We surveyed seven universities in the UK, and despite the ‘broadening’ of our data collection in an attempt that the selection of students would be representative of their universities, our sample remains a convenience sample. Nevertheless, convenience samples are not uncommon in student surveys in Hong Kong [85], USA [86], or Australia [87]. Similarly, in the USA, universities and colleges self-selected themselves to participate in the American College Health Association National College Health Assessment survey [88]. Recall bias, sociability and social desirability could influence self-reported data. As the questionnaire was compiled from different published sources, we assessed some variables by single item measures and other variables by scales that had been validated by others. However, the necessity of a general student health survey to be conducted within a short time in classes, makes the use of in depth measures for each health factor unfeasible. Hence, we did not investigate other forms of PA that include work-related, transport-related and domestic-related activities. Nevertheless activities performed during leisure-time are generally more enjoyable and undertaken by personal choice.