The data from the current study were drawn from the "Romsås in Motion" project. The project design, baseline results and main outcomes have been reported elsewhere 1920 (Lorentzen, Ommundsen, Jenum, & Holme, submitted). Pre- and post- intervention surveys in two suburban communities in Oslo in year 2000 and 2003 included a physical examination comprising height and weight measurements, the collection of venous blood samples as well as a main questionnaire (Q1) assessing information on health status and health-related behaviours. In addition, a specially designed questionnaire concerning physical activity (Q2), comprising a measure of the stages of change in physical activity 21 and several measures of theory-based possible psychosocial mediators of change in physical activity 222324, was handed out at the survey site and returned by mail. All instruments underwent translation and back-translation procedures, and were pilot-tested on a small sample. At the post-intervention survey, participants in the intervention district also completed a third questionnaire (Q3), also pilot-tested on a small sample, assessing exposure to and participation in specific intervention components 3. Questionnaire data and Body Mass Index (BMI) were used in current analyses. Participants provided their written informed consent at both surveys. The Regional Ethics Committee and The Norwegian Data Inspectorate approved the study protocol.

All individuals between 31 and 67 years in the intervention community were invited to participate in the baseline health survey (n = 2954). Of these, 1497 attended (50.7%), and 1204 (80.4% of the attendees) provided data on the outcome variables from Q2 (stages of change in physical activity and psychosocial influences). Based on registry information from Statistics Norway of the invited cohort, the attendees to the baseline survey had a slightly higher socio-economic status than the non-attendees 20. Furthermore, the proportion of participants of non-Western origin who provided baseline outcome data (16.4%) was lower than for those not attending the baseline survey (26.6%, p < .001) or providing these data.

Of the attendees, 1089 were alive and still living in the Oslo area in 2003, and were re-invited to follow-up assessments. Of these, 755 (69.3%) attended and 616 (56.6%) provided follow-up data on outcome variables from Q2. Of these, 603 answered at least one question from Q3, and were included in the analyses of the present study, representing an attrition rate from baseline of 44.6%.

The theory-based intervention aimed at generating a forward transition in stages of change in physical activity by favourably influencing the potential psychosocial mediators; social support, self-efficacy, attitudes, perceived control, and physical activity identity. The multi-component population-based intervention (Figure 1) comprised four main strategies:

Differences in baseline frequency distributions and means by population subgroups were investigated by Chi-square statistics and independent samples t-tests, respectively.

According to Baron and Kenny 13, several steps are required to demonstrate a mediation effect: 1) the intervention affects the outcome, 2) the intervention affects the hypothesized mediator, 3) change in the hypothesized mediator is associated with change in the outcome, and 4) the relationship between the intervention and the outcome is attenuated when controlling for change in the hypothesized mediator. The higher the attenuation of the effect, the greater is the potency of the mediator 35. As the stages of change in physical activity construct previously had shown linear relationships with several independent psychosocial variables 313637, also in the total intervention study sample (Lorentzen et al., submitted), linear multiple regression analyses were considered appropriate.

To test criteria 1 and 2, change in stages of change in physical activity and changes in psychosocial variables were regressed on variables expressing exposure to and participation in (yes/no) the different intervention components. These analyses were performed separately for each intervention component and each outcome. When both criteria 1 and 2 were met, a third regression analysis was conducted to test criteria 3 and 4, where change in stages of change in physical activity was regressed on both exposure and change in the hypothesized mediator.

To test for moderating effects by gender, age, ethnicity, education and BMI in the relationships between exposure to or participation in intervention components and change in stages of change in physical activity (criteria 1), stepwise hierarchical regression analyses with change score in stages of change in physical activity as the dependent variable were performed. The baseline level of the moderator variable was entered in step 1, exposure to or participation in the intervention component was entered in step 2, and in step 3 the interaction term of exposure to or participation in the intervention component × the baseline moderator variable was entered. This procedure was repeated separately for each of the potential moderators and for each of the intervention components that related significantly to change in stages.

Change scores (post test – pre test) of stages of change in physical activity and of each potential psychosocial mediator were used in the regression analyses. The significance level for all analyses was set at P < .05. Although the risk for making a type I error (incorrectly declaring a statistical significance) increases with multiple testing, the significance level was not adjusted for multiple comparisons for reasons outlined by various authors 3839. The number of cases available for each analysis varied due to missing data. All analyses were performed with the Statistical Package for the Social Sciences (SPSS) (version 13.0).

Baseline socio-demographic characteristics, distribution in stages of change in physical activity, and level of psychosocial variables for the sample are presented in Table 3. The majority of participants were of Western origin (91.2%), mean BMI was 27.1 kg/m², more than 60% of the participants reported that they were inactive or irregularly active (Precontemplation, Contemplation, and Preparation stages), and mean values on evaluative and affective attitudes towards physical activity were high. Those who were included in the present study were older (p < .001), had higher education (p < .01), and were more likely to be of Western origin (p < .001) than those lost to follow-up (see Table 3).

95.3% of respondents reported that they had noticed the "Romsås in Motion" project. Exposure to the different intervention components (Figure 1) varied a lot (Table 1). For example, 92.7% had heard about the physical fitness test, 75.8% had seen project brochures about physical activity, and 41.7% had seen the "Walk the stairs"-poster. Actual participation in intervention activities varied considerably, from 50.8% (at least once accomplished the test of physical fitness) to 1.5% (received a Green prescription from their physician).

Subgroup analyses (Table 1) revealed that, compared to their counterparts, a significantly higher proportion of Westerners, people with high BMI, and physically active persons at baseline had noticed the project and had been more exposed to and participated more in some intervention components. The latter also held true for women and persons aged 50 years or more.

As indicated by the β-weights in Table 4, participation in walking groups and aerobic exercise groups, having used the walk path, and seen the "Walk the stairs"-poster were significantly related to forward transition in stages of change in physical activity. Interaction analyses demonstrated gender (change in R² of step 3 p < .05) and ethnicity (change in R² of step 3 p < .01) interaction effects with respect to using the walk path. The relationship between having used the walk path and change in stages of change in physical activity was significant for women (β = .21, p < .01, n = 255) but not for men (β = -.01, p = .88, n = 185), and for Western persons (β = .15, p < .01, n = 407), but not for non-Westerners (β = -.29, p = .09, n = 33).

For all the intervention components which related to forward transition in the stages of change in physical activity, indication of psychosocial mediation was revealed. First, changes in all psychosocial variables related with the actual intervention component (step 2 in the Baron & Kenny formula) related significantly and positively to change in stages of change in physical activity (all p's<.05) (step 3 in the formula). Additionally, the β weights indicating the relationship between exposure to intervention components and change in stages (step 1 in the formula) were reduced when controlling for changes in the actual psychosocial variables (step 4 in the formula). Thus, the positive relation between having participated in walking groups and change in stages of change in physical activity was partly mediated by favourable changes in support from friends (β reduced .025 from step 1 to step 4 in the formula) and perceived control (reduction β = .029). The positive relation between having participated in aerobic exercise groups and change in stages was partly mediated by favourable changes in support from family (reduction β = .047) and from friends (reduction β = .06), self-efficacy faced with psychological barriers (reduction β = .056), affective attitudes (reduction β = .057), perceived control (reduction β = .069), and physical activity identity (reduction β = .056). The positive relation between having seen the "Walk the stairs"-poster and change in stages was partly mediated by favourable changes in support from friends (reduction β = .019), perceived control (reduction β = .064), and identity (reduction β = .029). Due to interaction concerning the walk path, mediation analyses for this specific intervention component were performed only for women and for Westerners. For women, social support from friends and perceived control mediated the observed significant relation between having used the walk path and change in stages of change in physical activity. For Westerners, support from friends, self-efficacy in the face of psychological barriers, and perceived control mediated this relation (see Table 4).

Variations in exposure and participation rates between different intervention components were expected due to the gradual introduction (Figure 1). The higher exposure than participation rates are consistent with the Persuasion Communication model postulating many steps in the process between initial exposure to a communication message and actual change in behaviour 3. Although differences in intervention programs, their duration and measurement methods make direct comparisons of results difficult, the level of intervention exposure found in our study is comparable to that of another study using a mass-media based multi-component intervention 40, promoting walking among sedentary 50- to 65-year-olds. Many study participants (32–90%) had been exposed to information through various media channels, but less than 5% reported seeing or participating in any of the public health education programs.

The significantly higher proportion of women compared to men who reported having participated in aerobic exercise groups is in accordance with other studies 18, although we did not find gender differences concerning participation in other intervention group activities. Not surprisingly, the non-Westerners were more difficult to reach with information messages than Westerners, despite the efforts made to reduce language and cultural barriers. Although exposure and participation rates in most intervention components were comparable for people with different educational status and BMI level, the higher participation rates of persons with low education in walking groups and of persons with high BMI at lectures are encouraging as these groups are considered hard to reach in health promoting work 1. Overall, high proportions of initially inactive persons were reached by intervention components, although less than among those who were physically active at baseline. This is in line with previous findings suggesting that present physical activity behaviour is related positively to past physical activity behaviour 24.

Findings suggest that having participated in walking and in aerobic exercise groups, having used the walk path, and having seen the "Walk the stairs"-poster were the most effective in moving intervention participants forward towards a higher stage of change. The sizes of the beta values suggest that taking part in aerobic exercise groups may have contributed the most. Our results are comparable with previous research. For example, several reviews have found support for the effectiveness of point of choice prompts to encourage stair use rather than escalators or lifts 12. Results concerning the walking groups fit nicely with the current emphasis on walking within public health interventions. Indeed, due to its acceptability and accessibility, walking groups seem particularly attractive among populations with low levels of physical activity 41.

Having used the walk path seemed to favourably influence stage movement in women but not in men, and in people of Western origin, but not in non-Westerners. The walk path passes by apartment buildings, a shopping centre, and public transportation terminals. Hence, the result may reflect that women and Western inhabitants used this path more in the purpose of exercising, and that men and non-Western people to a higher degree used the path as a means of transportation for accomplishing everyday tasks, which may not have influenced their level of motivation for participation in physical activity or their perceived levels of physical activity. No other moderating effects were found. Apparently, participating in walking and aerobic exercise groups, or seeing a "Walk the stair"-poster, seems equally able to influence stage movement, irrespective of gender, age, ethnicity, education level or BMI group.

The results of this study correspond well with and complement previous mediation analyses on the same material, indicating that several of the same psychosocial factors acted as mediators of the favourable intervention effect on stage movement (Lorentzen et al., submitted). The findings suggest that specific intervention components may have triggered specific psychosocial factors to influence the forward stage transition (Lorentzen et al., submitted). For instance, participating in aerobic exercise groups seems to play an important role in the previously observed increase in perceived support from family in the intervention group (Lorentzen et al., submitted). The latter is in line with former research emphasizing the great opportunity of gaining social support for physical activity behaviour by attending exercise groups 42.

As expected, all intervention components that significantly related to change in stages of change in physical activity seemed to partly have worked through increasing participants' general perceived control over physical activity behaviour 2232. Enhancement of peoples' perception of "low-treshold" physical activity opportunities in the local environment may have increased their perceived control over physical activity. This line of reasoning fits well with socio-ecological models stating that the environment may promote and maintain health-enhancing physical activity 43.

Several factors may explain why not all intervention components influenced potential psychosocial mediators of physical activity behaviour change or stage movement. First, a couple of these components were implemented rather late in the intervention period (Figure 1). Second, the low participation rates in some of the components resulted in low power for these groups in statistical analyses, making it difficult to obtain significant relationships. Indeed, similar intervention strategies have been found effective in increasing physical activity in previous research. For instance, comparable with the Green prescription, several studies have demonstrated that primary care physicians may play a significant role in the promotion of physical activity 44. Hence, more research is needed to rule out the potential effect of these components.

The main strengths of the present study include the measurement of indications of effect of specific intervention components on change in stages of change in physical activity 2, combined with the examination of numerous theory-derived potential psychosocial mediators of such changes 7. Second, the investigation of potential moderating factors in these analyses also provides important information 6.

Inherent in the present study are also several methodological limitations. Foremost, process evaluation data were retrospective and self-reported, increasing the possibility for both recall bias and social desirability bias 45. For instance, people who have changed their behaviour after being exposed to specific intervention components may be more likely to remember the components than people who did not change their behaviour. Recall and social desirability bias may also apply to the self-reported stages of change in physical activity and psychosocial factors. For example, Ronda and colleagues 46 reported that more than 60% of those insufficiently active overestimated their activity level. The results of the current study may also be limited by the lack of a comparison condition. The non-controlled, retrospective evaluation design reduces the ability to make certain conclusions about the effects of the specific intervention components on psychosocial influences and stages of change. However, several of the psychosocial variables positively related to exposure to and participation in different intervention components were the same as those found to be positively affected by the whole intervention strategy package in analyses applying a prospective controlled design (Lorentzen et al., submitted). This, together with the significant favourable intervention effect on forward stage transition revealed in the same study, support the main findings of the present study.

For better understanding the effects of the present intervention, more detailed process evaluation information, such as the reading of brochures, the receipt of information at stands, or reasons for participating or not in intervention components, could have been assessed 3. Low response rates at baseline (40.8 %) and at follow-up (55.4 %) may also have biased the results. As attrition analyses indicated that baseline and follow-up samples comprised people with slightly higher education and more Westerners compared to those not attending or not providing relevant data at these assessment time-points, the results may not be generalized to the same extent to these subgroups. Further, since successful promotion of physical activity is recognized as more challenging in groups of low socio-economic status and groups of ethnic minorities 1, the results observed in the present study might have been less positive if such selection effects had not taken place. However, no differences were found on baseline values of other socio-demographic characteristics, BMI, distribution in stages of change in physical activity, or psychosocial mediators between those who completed both assessments and those lost to follow-up, making the present sample fairly representative of those providing baseline data. These findings also indicate that those who provided data for the present study did not have a higher baseline psychological readiness for increases in physical activity than those who did not. However, people who have become more physically active as a result of the intervention might still be overrepresented, and the present sample was characterized by a relatively high baseline physical inactivity rate and BMI score, which further limits generalization of results.

The results should also be interpreted with caution as some say that the stages of change construct should not be used as a proxy for physical activity change, as movement between some stages do not involve change in physical activity level 47. However, the significant positive relationship found between the stages of change in physical activity and strenuous physical activity in leisure and commute time in the main study (Lorentzen et al., submitted) lends support for the use of the stages of change construct as an outcome measure. Finally, the lack of a precise definition of regular physical activity may have reduced the reliability and validity of the stages of change measure, as Martin-Diener and colleagues 48 found that an individual can be in different stages of change depending on the definition of the behaviour.