Participants and procedure
The participants were recruited with the aim of attaining a reasonable representation of the adult active commuters in the inner urban and suburban areas of Greater Stockholm during the recruitment period. Active commuters constitute a small group within the general population and therefore it was not possible, in practical terms, to recruit a sufficient number of participants from a random population sample. We therefore recruited participants by advertising in two large morning newspapers in Stockholm (Dagens Nyheter and Svenska Dagbladet) towards the end of May and early June 2004. Inclusion criteria were: (a) being at least 20 years old; (b) living in Stockholm County, excluding the municipality of Norrtälje; and (c) walking and/or cycling the whole way to one's place of work or study at least once a year. In the invitation to participate, we emphasized that people with short commuting distances were also welcome to participate. The reason for including people with less frequent active commuting behaviours, as well as with short route distances, was to include a wide range of commuting behaviours.
The advertisements resulted in 2,148 individuals volunteering to take part. We posted a first questionnaire, called the Physically Active Commuting in Greater Stockholm Questionnaire (PACS Q1; for further description, see below), to the participants in September 2004. The response frequency was 94% (n = 2010). During the peak bicycle-commuting period of the year, in May 2005, a second questionnaire, the PACS Q2, was sent to 1978 participants. The response frequency was 92% (n = 1819). Both questionnaires were sent home to each participant together with a prepaid return envelope. A maximum of three reminders were sent out. No incentives were provided for participation. We excluded some participants in the second round because they did not meet the inclusion criteria, or did not wish to participate in the second part of the study. The participants were bicyclists, pedestrians or dual-mode commuters, i.e. individuals who sometimes walk and sometimes cycle. They commuted in the inner urban or suburban - rural areas of Greater Stockholm, or both of these areas. Since these areas represent distinctly different environmental settings (for details, see [7
]), we believe that it is of clear importance, at least initially, to study them as separate entities. In this study we have therefore only used data on bicycle commuting in the inner urban area. After cleansing and editing the data, 827 participants (women, n = 499, 60%) were included in the analyses. We were concerned about whether or not bicycling and ratings of route environments in two different areas, as compared to only one area, would affect the rating levels. Therefore, in a previous study [7
], partly based on the same participants as those in this study, we compared ratings of inner urban route environment between those who bicycle commuted in both inner urban and suburban - rural areas (n = 555) and those who bicycle commuted in only an inner urban area (n = 272). Overall, the results indicated only few and small differences between the groups. Hence, in this study we combined the two groups. For further descriptive characteristics of the participants, see Table .
Descriptive characteristics of participants (n = 816-826)
We were also concerned about the representativity of the advertisement-recruited participants. We therefore, in a previous study [7
], compared the ratings of route environments between advertisement- and street-recruited participants. The street recruitment strategy was considered to represent the population of active commuters at that period of recruitment with greater certainty than the advertisement strategy. Overall, the results indicated a good correspondence between the advertisement- and street-recruited participants' ratings. For example, the sex-neutral mean values for the different items for the different groups were gathered along the line of identity for both urban and suburban areas, and the Pearson correlation coefficients were 0.96-0.98 [7
The Ethics Committee of the Karolinska Institute approved the study. The participants gave their informed consent.
The physically active commuting in Greater Stockholm questionnaire (PACS Q)
The PACS Q1 and PACS Q2 are self-administered questionnaires in Swedish, based on self-reports. They include 35 and 68 items, respectively, comprising descriptive characteristics of participants and different aspects of active commuting. The PACS Q2 includes the ACRES.
Measure of descriptive characteristics
Data on sex, age, weight, height, employment and number of bicycle-commuting trips per month were obtained from the PACS Q1. The body mass index (BMI) was calculated by dividing body weight by height squared (kg·m-2). Active commuting trips per year were calculated by adding each month's average trip frequency per week and then dividing the sum by 12 to obtain values for an 'average week', which were thereafter multiplied by 52. Education levels, income, ethnicity, having a driver's licence, having access to a car, time leaving home to cycle to work and overall physical and mental health were obtained from the PACS Q2 (Table ).
The active commuting route environment scale (ACRES)
The ACRES consists of 18 items for the assessment of bicyclists' perceptions of their self-chosen commuting route environment, potentially associated with active commuting. A more detailed description of the development of the ACRES, its items and its validity and reliability has been reported elsewhere [7
]. The ACRES was characterized by considerable criterion-related validity and reasonable test-retest reproducibility.
Each item considers the inner urban area of Stockholm, the capital of Sweden, and the suburban as well as rural areas surrounding it, within Stockholm County, separately. The questionnaire instructions include a drawn map that distinguishes the inner urban area from the surrounding areas, see [8
]. The participants were asked to differentiate between their experiences when their active commuting route is in the inner urban area and when it is in the surrounding suburban as well as rural areas (Figure ). All items have two identical parallel response lines. One line refers to the inner urban area and the other to the suburban as well as rural areas. The separation between these parts was based essentially on the fact that they constitute different environments: the inner urban area is a dense urban setting with blocks placed in a grid-like streetscape, typical of European cities, whereas, with very few and small exceptions, this is not the case in the suburban-rural areas. For a detailed description of each area's environmental features, see [7
Figure 1 Example of an item from the Active Commuting Route Environment Scale (ACRES) for bicyclists. The participants were asked to differentiate between their experiences when their active commuting route is in the inner urban area and when it is in the surrounding (more ...)
To simplify understanding, the items for the assessment of bicyclists' perceptions have been divided into: (a) the physical environment; (b) the traffic environment; and (c) the social environment. The following items are included in the physical environment (see Table ): bicycle paths/lanes/roads (#11), greenery (#13), ugly or beautiful (#14), course of the route (#15), hilliness (#16), red lights (#17) and short or long (#18). They represent non-moving aspects. The following items are included in the traffic environment: exhaust fumes (#3), noise (#4), flow of motor vehicles (#5), speeds of motor vehicles (#6), speeds of bicyclists (#7), congestion: all types of vehicles (#8) and congestion: bicyclists (#9). They represent moving aspects. The following item is included in the social environment: conflicts (#10). It represents relationships between road users. All items are meant to operate independently. The remaining three items, namely, on the whole (#1), hinders or stimulates (#2) and traffic: unsafe or safe (#12), are regarded as outcome variables. All the other items are regarded as predictor variables believed to be potentially important for the outcome variables. The numbers specified in parentheses indicate the order in the questionnaire; see Table . In this study short or long and on the whole are not used.
The Active Commuting Route Environment Scale (ACRES) for bicyclists
The reason for not using the perception of short and long is that it refers to the whole trip distance and the fact that a significant portion of our participants do also cycle in the suburban area. The reason for not using the perception of on the whole is that this item is too general for the purpose of this study. On the other hand, given that this is an exploratory analysis of what constitutes the overall perception of whether a route environment is hindering or stimulating, we have included traffic: unsafe or safe (which normally is viewed as an outcome variable) as a predictor variable in model 2 in our analysis (see below). This is to check if there are indications that there might be an overlapping environmental basis for these two different outcome variables.
Fifteen-point response scales, with adjectival opposites, ranging from 1 to 15, corresponding to, for example, 'very low' and 'very high', are used, with the exception of one item. The item bicycle paths/lanes/roads has an 11-point response scale ranging from 0% (0) to 100% (10) (Table ). The 15-point response scales feature a numbered continuous line, i.e. whole numbers from 1 to 15, with number 8 as a neutral option in the middle, labelled, for example, 'neither low nor high' (Figure ).
In the questionnaire instructions, the participants are asked to recall and rate their overall experience of their self-chosen route environments based on their active commuting to their place of work or study during the previous 2 weeks. The reason for this is that we wanted them to have fresh perceptions. Individuals stating that they had not been cycling the last 2 weeks were therefore excluded. At no point are the participants informed about the intent of the ACRES.
The commuting route environments are located in the inner urban area of Stockholm, the capital of Sweden, in the centre of a metropolitan area with about 1.9 million inhabitants. This area constitutes the region's single core urban structure, with the centre situated where Lake Mälaren meets the Baltic Sea, thereby dividing the region into two main parts. The study area includes the city sections of 'Gamla stan' (the Old Town), Södermalm, Kungsholmen, Vasastan, Norrmalm and Östermalm (Figure ). This is a predominantly built-up area, with blocks in a grid-like streetscape. The age of the buildings varies. The Old Town is from medieval times, whereas most parts of the built-up environment are predominantly a result of the architectural styles from the end of the 19th and beginning of the 20th century, with most buildings about five storeys high. The newest part of the city centre is north of the Old Town. The original buildings here were torn down during the 1950s and 1960s, and today the area includes modernistic architecture, including a few skyscrapers. In 2005 the residential density of the inner urban parts of the study area was approximately 13 000 residents per square km [9
Figure 2 Aerial view from 2005 over the inner urban parts of Greater Stockholm, Sweden. The yellow line distinguishes the inner urban and suburban - rural parts. North is on the left of the image. For description of the characteristics of the study area, see Methods. (more ...)
The city has a number of waterfronts and islands, a number of both small and large parks, some alleys and esplanades. Most streets are void of trees or other forms of greenery. The natural landscape in the area is sediment-filled valleys as a part of the surrounding rift-valley landscape and raised archipelago landscape with eroded bedrocks after deglaciation. It is basically rather flat, but there are some dominant natural features such as, for example, part of an esker, rising 40 m above sea level in Vasastan, as well as a rather steep fault scarp in Södermalm. The road system also includes rather gentle slopes of infrequent moraine hills, normally not accounting for more than about 10-15 m of elevation. Two arterial highways pass through the inner urban area (Centralleden and Essingeleden), but cyclists or pedestrians come into very little contact with them. These are also the only roads, besides some tunnels, that do not permit cycling.
Questionnaire data were entered in the Statistical Package for the Social Sciences, version 19.0 (IBM SPSS Inc., Somer, NY, USA). All entered data from the PACS Q2 were checked for accuracy. Some participants were excluded, mainly because of incorrect or incomplete ACRES data. Participants with three or less missing ACRES values for cyclists were used for the following measures: (1) percentages and mean scores ± 1 standard deviation (SD), used to report the characteristics of the participants.; (2) the values of the ACRES items, presented as mean scores ± 1 SD; and (3) interrelations between the variables assessed with Pearson's correlation coefficient (r).
Simultaneous multiple regression analysis was chosen to explore associations between the outcome variable, hinders or stimulates, and the predictor variables exhaust fumes, noise, flow of motor vehicles, speeds of motor vehicles, speeds of bicyclists, congestion: all types of vehicles, congestion: bicyclists, conflicts, bicycle paths/lanes/roads, traffic: unsafe or safe, greenery, ugly or beautiful, course of the route, hilliness and red lights. Two models were run. In Model 1, traffic: unsafe or safe was excluded, and in Model 2, it was included as a predictor. The reason for including traffic: unsafe or safe; a variable that we normally regard as an outcome variable, was, as stated previously, its possible association with the outcome variable: hinders or stimulates. Only participants that had no missing values for any of the studied ACRES variables were used in the simultaneous multiple regression analyses.
Before running the simultaneous multiple regression analyses, linearity of the variables was assessed visually by means of scatterplots, boxplots and errorbars. All variables demonstrated reasonable linearity and were therefore used in the analyses. Furthermore, before the analyses, interrelations between the variables were assessed with Pearson's correlation coefficient (see Table ). The correlations between predictor variables were, in absolute values, r ≤ 0.68, indicating no problems with multicollinearity. In addition, multicollinearity was checked with the variance inflation factor (VIF). Both models' VIFs (all values ≤ 2.26, mean: 1.65) indicated no problem with multicollinearity.
Correlations between ratings of environmental variables (n = 818-827)
The top limit for inclusion of standardized residuals in the models was set to ± 4 SD, according to the sample size used [10
]. Possible extreme data cases were defined using Cook's distance. No extreme data cases could be defined using Cook's distance in either of the models (all values ≤ 0.03, mean: 0.001).
Sex (dichotomous categorical variable), age (continuous variable), education (categorical variable coded as dichotomous: university/university college or other lower) and income (categorical variable coded as three categories and used as a dummy variable: ≤ 25 000 SEK, 25 001-30 000 SEK or ≥ 30 001 SEK; SEK = Swedish crown/krona, year 2005: €1 ≈ 9 SEK; US$1 ≈ 8 SEK) were possible confounding variables. Before considering using them in the simultaneous multiple regression analyses, we assessed their individual contribution to the variation in the outcome variable using simple regression analyses for the sex, age and education variables, and simultaneous multiple regression analysis for the income variable. The results demonstrated no significant contribution or a very small significant contribution (age: R2 = 0.008). We have therefore chosen not to include these variables in the simultaneous multiple regression analyses.
The values from the simultaneous multiple regression analyses are presented as unstandardized beta coefficients (B) and their 95% confidence interval (CI), and partial correlation coefficients. Furthermore, the R square (R2) is presented for the overall models.
A statistical level corresponding to at least p ≤ 0.05 was used to indicate significance.