An often proposed strategy to improve public health through physical activity is to increase the amount of physically active transportation, for example, commuting by walking or bicycling (7,15). An additional benefit of such a behavioral change would be the lessening of the multifaceted environmental burden of the transport system.
Distance is of pivotal importance for understanding a variety of aspects of transportation. For example, health effects of physical activity are related to levels of induced energy turnover per week (8,12). Translated into physically active transportation, this correlate of health can be broken down to the variables of route distance, frequency of trips, and energy demands of different modalities of transport, for example, cycling or walking. Distances are also an important form of input data in cost-benefit analyses of investments in infrastructure for walking and cycling (14) and in the World Health Organization's Health Economic Assessment Tool for cycling (16). Furthermore, the prevailing physically active commuting distances are of interest in relation to the distribution of distances between home and work places within the population because this relation can reflect the potential of physically active commuting as a means of transportation in a population. Thus, there are many reasons for the need of accurate and reliable methods for determining distance in the context of physically active transportation.
Such methods need to be based either on distance measurements of the actual routes taken or on indirect methods shown to be able to mimic these routes or distances. After an extensive literature survey in the research fields of physical activity, exercise, and transportation, we have concluded that there is a lack of studies establishing a criterion method for distance determinations in physically active transportation. The issue therefore arose as to whether any of the commonly used distance estimation methods within the transport sector, although most often focusing on motorized vehicles, could be used. Examples of such methods are shortest or fastest route algorithms in geographical information systems (GIS), straight-line distance, or self-reported distance estimations (see, e.g., [1,3,11,13,17]). It is clear that straight-line distance or self-reported distance estimations are not reliable (see, e.g., ). GIS calculations are based on theoretical and uncertain approximations of the actual routes. Furthermore, it is a rule that the map information databases coupled to GIS systems do not take into account whether the routes can be used for walking or cycling nor do they normally include paths specifically destined for physically active transportation. However, even in cases where map information has included more detail on routes specifically available for cyclists, these systems have not been able to produce valid distance data in a context of physical active transportation (2). Indeed, shortest and fastest routes might be rational choices in the motorized transport modes, whereas other reasons seem to influence behavior in physically active transportation. Duncan and Mummery (6) traced routes taken by children to school with global positioning system (GPS) and found that they differed from those selected by GIS in shortest route analyses. The reason for this deviance in route choice seemed to be a calmer traffic environment along the self-chosen route. Similar results have been obtained by Aultman-Hall et al. (2) who studied adult bicycling commuters.
Thus, existing distance measurement methods within the transport sector do not seem to be appropriate in a context of physically active transportation. In principle, GPS-based route tracings and measures of distance are of interest for this purpose. However, in practical terms, they are quite elaborate and have been shown to sometimes produce raw data with different kinds of error, for example, signal noise (6,10). We have therefore chosen to illuminate whether measuring the distance of physically active commuting routes drawn on maps can be used as a part of a criterion method. This has been studied through checking the validity and reproducibility of a digital curvimetric distance measurement system, different manual measurement techniques, and the distances of routes drawn by commuting pedestrians and bicyclists of both sexes.
The study area is the center of the County of Stockholm, Sweden, a metropolitan area of approximately 1.9 million inhabitants. It includes commuting routes in the inner urban area of Stockholm and the suburbs surrounding it. The road systems in the former area are characterized mostly by a grid net streetscape, whereas in the suburbs, roads are less structured and are often influenced by the topography of the natural landscape.
Recruitment of participants.
Participants were recruited from cyclists and pedestrians as they slowed down at the top of four bridges or stopped at a traffic light by one arterial road. All contacts were made in central Stockholm between 7 and 9 a.m. in mid November 2005. In total, 589 response coupons were distributed and 214 coupons were returned in due time.
The inclusion criteria were as follows: minimum age of 20 yr, domicile in the County of Stockholm excluding the municipality of Norrtälje, and walk and/or cycle the whole way to work or study place at least once a year. In the information to the commuters, it was underlined that people with very short commuting distances were also welcome to participate in the study. Thus, the invitation to participate welcomed all types of behavior in commuting distance and trip frequency.
In order not to influence the results, the participants were, neither at the point of recruitment nor at the first distribution of maps and questionnaires, informed about the objectives of the study. At the second distribution of the same material, the participants were informed that the reason for duplicating the procedure was to value how certain the data were.
The study was approved by the ethics committee of the Karolinska Institute, and the participants gave their written informed consent.
The response rate in returning the two maps and questionnaires was 64.5%. Five respondents did not meet the inclusion criteria and were excluded from the study. For the remainder (n = 133), self-reported age, height, and weight (mean ± SD) were 44 ± 10 yr, 1.67 ± 0.06 m, and 63 ± 7 kg for the women (n = 56) and 46 ± 11 yr, 1.81 ± 0.07 m, and 79 ± 11 kg for the men (n = 77), respectively. Routes drawn were validated using a subgroup of 19 participants who were comparable with the whole group concerning age, height, and weight.
All of the participants lived and worked either in the inner urban area or in the suburbs. Thus, no one lived or worked in rural areas. Their average self-reported number of commuting trips per week, for each month of the year, ranged between approximately 4 and 9, if the dominant vacation month of July was left out (n = 121). There were small variations between the female and male bicyclists as well as the pedestrians. Seventy-six percent of the participants (n = 127) were educated at university level.
Questionnaires, maps, and instructions.
From the home and work or study place addresses stated in the response coupon, individually adjusted copies of maps were sent by post to the participants in November to December 2005 together with a missive and a questionnaire. The maps used were copied from the main telephone book in the County of Stockholm (scale 1:25,000). A second letter, with the same material and tasks as in the first letter, was, with some few exceptions, sent to the participants 2 wk after they had returned the first questionnaire and map. The median time between the participants' return of the first and second maps was 21 d (range = 12-59 d).
The participants were instructed on how to draw their most common commuting routes from home to work and vice versa on the map. Participants with two or more work places were asked to pick the one they spent most time at and, in the case of equal time spent, to choose one of them. They were then asked to walk or cycle their route once, noting their route choice and the street names, before filling it in on the map. For bicyclists, the route from home to work place was to be marked with a plain line (Fig. 1); for pedestrians, with a line with crosses on. The route from work to home, if not the same as the route to work, was to be marked with a line with circles on for bicyclists, and with triangles on for pedestrians. The location of their homes was to be marked with a "B," and schools/nursery schools with an "S," if dropping off children was part of the trip between home and work. The precise destination point at their work place was to be marked with a filled square. Finally, the respondents were asked to carefully mark their routes in places outside the printed street grid network, such as tunnels and parkways.
Ninety-eight percent of all the maps were drawn according to the given instructions. With the answers given in the questionnaires and response coupons, the remainder of the route markings could be interpreted securely.
Validity and reproducibility of distance measurement tools and simulated distance measurements.
To create each individual map, the original map had to be copied in two steps. The original scale of the map was conserved in the copies.
We found great differences in reproducibility in a pilot test of different analog and digital distance-measuring tools. The tool with the best result, a digital curvimetric distance measurer (Run Mate Club; CST/Berger, Watseka, IL), was chosen for thorough control of both validity and reproducibility.
This was performed by using a straight 250-mm-long line and measuring it 20 times. The average relative difference in length was +0.4 ± 0.2% (P ≤ 0.001). In the pilot test of different equipment, we noted that changes of measurement directions were often critical points in introducing errors. Therefore, the validity of the tool was also tested on a right-angled triangle with the lengths of the sides being 30, 40, and 50 mm. Two different manual measuring techniques were used: lifting the distance measurer and rotating it in the air when arriving at an angle or turning it while still on the map. Each technique was measured 10 times, each time consisting of five rounds along the triangle. Thus, each time, a total distance of 0.6 m with 14 distinct changes in directions was measured. The average relative differences with the lifting technique were +0.2 ± 0.4% (n.s) and for the turning technique +1.0 ± 0.5% (P ≤ 0.001).
A technical assistant performed all map measurements using the turning technique. The validity and reproducibility of this applied map measuring were tested in a mimicked commuting route setting. A line with a total length of 319 mm and consisting of four right angles, two flat angles of 120° and 135°, and one sharp angle of 35° was used. It was tested 10 times before and 10 times after 2.5 h of map measurements. No difference in accuracy was noted between these occasions, and the average value of the 20 measurements differed −1.3 ± 0.6% (P ≤ 0.001) from the real value.
Measurements of route distances.
The distances of the map-drawn commuting routes from home to work were measured twice. A third measurement was undertaken if differences between the first two values were higher than 4.7 ± 2.3 mm (n = 23), that is, 1.7 ± 1.7% of the first measured value. The two closest values of the three were chosen, which resulted in two values differing on average 1.1 ± 1.0 mm corresponding to 0.4 ± 0.5% (n.s.). The quality of the technical assistant's work was controlled by one of the authors checking the route distance measurements of 30 randomly chosen maps. Any deviances found were within the error of the method.
Validity and reproducibility of markings of origin and destination points on maps.
As mentioned previously, a control of the maps was performed to ensure that all marked routes agreed with the inclusion criteria; the whole way from home to work or study place, and not, for example, to a bus or train station. For this purpose, the starting points and destinations marked on the maps were compared with the home and work or study place addresses given by the respondents. Three address-geocoding systems on the Internet (Google Earth (version 4.0), www.eniro.se, and www.hitta.se) were used to locate the respondents' street addresses on digital maps.
This control also formed the basis for evaluating the validity and reproducibility of the markings on the maps. Deviances between address-geocoding systems and the respondents' markings of 3 mm or less were deemed unnecessary to report. Deviances of more than 3 mm are reported and their sizes are stipulated. The same criterion was used for the reproducibility of the markings of origin and destination points between each individual's first and second maps.
GPS test of the validity of map-drawn routes.
A subgroup (n = 49) was selected from all the participants because of their good reproducibility when drawing their commuting routes. Lateral route marking deviations in this group were 3 mm or less. The reason for forming this subgroup was to validate their map-drawn routes with a GPS. However, pilot tests with other GPS had shown that GPS tracings sometimes displayed routes not taken, such as crossing buildings. Such errors can be caused by losses of satellite signal reception created by high buildings or routes through tunnels. Thus, it was considered important that the subgroup studied could be anticipated to take the drawn commuting routes. In such cases, we would attain a validity control of the GPS used.
For this purpose, a letter was sent to the participants of the subgroup in spring 2007 asking if they were interested in participating in a new aspect of the study. It included a copy of a map with their previously drawn commuting route. An inclusion criterion was that they still had the same route to work as previously. Of the 39 responding persons, 6 women and 13 men met the criterion and were willing to participate.
These participants were equipped with a GPS (SPI10; GPSports Systems Inc., Canberra, Australia). The GPS was put in a harness on the upper back to enable a good satellite reception. The walking or bicycling trip started when the GPS had contact with at least four satellites. The logged coordinates from the GPS were transferred to a GPS mapping software (OziExplorer version 3.9; Ozi-Explorer, Inc, Brisbane, Australia) equipped with a Stockholm county map ("Gröna kartan," scale = 1:50,000; Lantmäteriverket, Gävle, Sweden, 1999). Tracks formed from the coordinates were printed out in the same scale as the paper maps and were then compared with the drawn routes. If lateral deviances of more than 3 mm between the drawn routes and the GPS tracks were found, the length of the deviating parts was measured. The GPS was not used for any distance measurements. Instead, all distances were measured with the digital curvimetric device.
Given the fact that we observed unexpected deviations between GPS tracings and drawn routes on the maps (see Results), it was considered important to ensure that these deviations were real. One of the authors therefore cycled seven different determined routes in urban environments and seven in suburban environments, with a total distance of 43 km. This procedure was repeated to check on the reproducibility. With the exception of the medieval part of Stockholm, which has very narrow streets, the GPS tracings were easily matched with the actual routes. Thus, we are confident that the deviations reported in the Results are valid.
Values are presented as mean ± SD unless otherwise stated. All statistical analyses were performed using Statistical Package for the Social Sciences version 15.0 (SPSS Inc., Chicago, IL). The validity and reproducibility control of the digital curvimetric distance measurer was undertaken with Student's paired t = test to compare given distances with repeated measures of them. An intraclass correlation analysis of test-retest in distance values of drawn commuting routes was performed. The possibility of significant order effects in the test-retest measurements of route distances were analyzed for each group of gender and transport modality (pedestrians and bicyclists) using Student's paired t-test. The differences in test-retest route distance values between the four different groups were compared using a two-way ANOVA for independent samples. The typical error of the method (9) was calculated for each group and for all participants. Differences in measured distances between routes drawn and the route choice indicated by the GPS tracings were analyzed with Student's paired t-tests. A statistical level of at least P ≤ 0.05 was considered significant.
Reproducibility of Map-Drawn Route Distance Measurements
The test-retest intraclass correlation coefficient was 0.999 (P ≤ 0.001; Fig. 2). No significant order effects between test and retest occasions were noted within any of the groups (Table 1). The differences for each group did not differ significantly between the groups, and the average for all participants was 12 ± 229 m. The test-retest differences were in absolute terms of the same order of magnitude throughout the studied range of distances (Fig. 2). The typical error of the method for each group is presented in Table 1; and the typical error of the method for the whole group was 162 m.
Validity of Routes Drawn and Its Effects on Distance Measurements
A pertinent issue is whether routes drawn and measured concerning distance are the actual routes taken and, if not, whether deviances between routes drawn and routes taken would introduce any measurement error of significance. For this purpose, we checked the validity and reproducibility of all participants' markings of origin (home) and destination points (work or study place) in relation to their stated addresses for these points. We used this measure as an indicator of exactness of the route drawings at two points and as an indirect indicator of route drawing exactness in general. Furthermore, we also traced a subgroup of 19 persons with GPS as they walked or cycled their actual commuting routes. The results are described in the following text.
Validity of markings of origin and destination points in relation to stated addresses.
Ninety-six percent of the map markings of origin points (home) agreed (≤3-mm deviation; 3 mm in maps corresponded to 75 m in reality) with the stated addresses. This applied to all three address-geocoding systems used to localize the addresses stated by the participants. The corresponding figure was 95% for the destination points. The mean values of the deviations divided between all 133 individuals were 0.21 mm for the origin points and 0.62 mm for the destination points (Table 2), which correspond to less than 0.1% and 0.3%, respectively, of the total measured average route distance length (6.8 km).
Reproducibility of markings of origin and destination points on maps.
Ninety-six percent of the origin markings were reproduced within 3 mm from each other. The corresponding value for the destination points was 88%. The mean values of the deviations divided between all 133 individuals were 0.20 mm for the origin points and 1.37 mm for the destination points (Table 3); these correspond to less than 0.1% and 0.6%, respectively, of the total measured average route distance length (6.8 km).
Validity of map-drawn route distance measurements.
The first step in this validity check was to compare the routes drawn on the maps with the tracks recorded by the GPS. Of the 19 participants, 13 had identical or almost identical GPS tracks (≤3-mm lateral deviation) compared to the map-drawn routes. In six participants, there were lateral deviations of more than 3 mm. The lengths of the deviating sections of the routes varied between 1% and 12% (7.0 ± 4.6%) of the total lengths of drawn routes. The main characteristic of the deviations was that GPS tracks were located on streets parallel to those drawn on the maps. The effect of these route deviations on the total distance measured on the route-drawn maps was 1.21 ± 1.60% (n.s.) in the six cases with deviations, and in relation to the whole group, this figure decreased to 0.38 ± 1.02% (n.s.), with a 95% confidence interval of −0.18% to 0.94% of the whole route-drawn distance.
To our knowledge, this is the first study that establishes that when physically active commuters draw their commuting routes on paper maps, they create a valid and highly reproducible basis for measurements of actual route distances. Several criteria have to be met in establishing this as a criterion method. They will be discussed here.
First of all, one must be certain that the scale of the original printed map is not changed through copying it. Then, the validity and reproducibility of the chosen distance-measuring system must be controlled both per se as well as during actual measurements of drawn commuting routes. How this was controlled is described in detail in the Methods section, with high validity and reproducibility demonstrated.
The very close resemblance in the test-retest values of drawn route distances mirror the fact that there was no order effect on the measurements. This was the case for all the groups studied. Neither was there any indication that differences in commuting distances played a role in this interpretation.
So far, the conditions described for the chosen method are excellent for establishing a criterion method. However, a crucial question is whether we can be sure that the drawn routes are the actual ones taken. If so, we are dealing with a criterion method. If not, then it is important to determine whether any deviances between the drawn maps and the actual routes are of any consequence for the distance measurements.
Checks for this were undertaken. First, the validity and reproducibility of origin and destination points marked on the maps were studied and found to be good. This is also an indirect indication that the participants had drawn the actual routes with care. The last control dealt with whether the routes drawn represented the actual routes taken and whether any deviances would affect the estimated distances. On the basis of a comparison between GPS tracings of commuting and map-drawn routes of a subgroup (n = 19), any deviances were found to play an insubstantial role for the route distance measurements.
The fact that the GPS tracings indicated that, in some cases, there might be deviances between actual routes and the routes drawn on maps warranted a detailed control for any deviances in route markings done by the 133 participants. In the test-retest of two consecutive route drawings on map sections with marked paths, streets, or roads, we found that 46% of the participants had, on average, 1.4 lateral deviances corresponding to a distance of about one block (Stigell and Schantz, unpublished observations). This mirrors the deviances indicated by the GPS tracings, which were shown to only have a very minor effect on distance. Indeed, the high reproducibility in the determined route distances in this study demonstrates that the lateral route marking deviations observed are unimportant for the distance measurements. Furthermore, because the sizes of the deviations between the drawn routes and between drawn routes and GPS tracings are of the same magnitude, it seems reasonable to assume that this reflects the size of deviations in potentially existing differences between both drawn routes and actual route taken within the whole group of participants. This will lead to the same conclusion in insignificant effects on the route distance measurements. Thus, taking all mentioned points into consideration, we are confident that a route-drawn map is both a valid and highly reproducible basis for route distance measurements in physically active commuting.
To be able to judge the external validity of our results, it is of value to state that the distributed paper maps were copied from the telephone book, which is a well-known artifact in the studied region. At the same time, it should be mentioned that this map is colored, whereas the respondents received a black and white copy. Another aspect of using the map was that the participants could easily verify their route by looking at the street names, which, to a high extent, could also be found on the map. This was a suggested strategy to the participants before drawing their routes. However, we do not know to what extent those instructions were applied. Could the good results be due to a specific selection of respondents with competence in map route drawing? We do not believe that this is the case for the following reasons. Of all the information sheets handed out, some pedestrians and bicyclists probably realized that they did not meet the inclusion criteria. Given this possibility, the response rate obtained was still 36% of everyone who received an invitation to participate in the study. Because we do not know who the nonrespondents were, we could not send any reminder letter. We have therefore compared our response rate with population surveys performed by Statistics Sweden, a national government authority for official statistics, after the first letter has been sent out. Their response rate is normally 20-30% (Alf Asplund, personal communication). It should also be noted that Swedes are trained to read maps through orienteering during physical education in school. Orienteering is also part of the obligatory military training, which the majority of the male participants would have completed. Also, commuting is a habit for our participants, so they are well acquainted with their routes. The last comment on the external validity dimension of our results is that the participants represented a well-educated cohort of the population; 76% of them were educated at the university level. The validity and reproducibility of this method in physically active transportation for other reasons than commuting (17), as well as in other settings, including cultural and geographical ditto, therefore deserve specific studies.
In conclusion, we have demonstrated that letting physically active commuters draw their commuting routes between home and work or study place and measuring these distances with a digital curvimetric distance-measuring device is a measuring method with both high validity and reproducibility. We have thereby established a potential criterion method that is useful per se and as a basis for future studies in the expanding field of physically active transportation research.
The authors are grateful for the financial support of The Research Funds of The Swedish Road Administration, The Public Health Funds of the Stockholm County Council, The Swedish National Centre for Research in Sports as well as from GIH - The Swedish School of Sports and Health Sciences. The authors are also grateful for the superb technical assistance of Mrs. Cecilia Schantz-Eyre and to Dr. Johnny Nilsson for supplying the GPS instrumentation. Finally, the authors thank MPT Jane Salier-Eriksson for her careful assistance in scrutinizing the use of the English language.
No conflicts of interest exist. The results of the present study do not constitute endorsement by the ACSM.
1. Apparicio P, Shearmur R, Brochu M, Dussault G. The measure of distance in a social science policy context: advantages and costs of using network distances in eight Canadian metropolitan areas. J Geogr Inform Decis Anal
2. Aultman-Hall L, Hall FL, Baetz BB. Analysis of bicycle commuter routes using geographic information systems: implications for bicycle planning. Transp Res Board Transp Res Rec
3. Chalasani VS, Denstadli JM, Engebretsen O, Axhausen KW. Precision of geocoded locations and network distance estimates. J Transp Stat
4. Denstadli JM, Hjorthol RJ. Testing accuracy of collected geoinformation in the Norwegian Personal Travel Survey-experiences from a pilot study. J Transp Geogr
5. Duncan MJ, Mummery WK. GIS or GPS
? A comparison of two methods for assessing route taken during active transport. Am J Prev Med
6. Duncan MJ, Mummery WK, Dascombe BJ. Utility of global positioning system to measure active transport in urban areas. Med Sci Sports Exerc
7. EU Commission. Promoting healthy diets and physical activity: a European dimension for the prevention of overweight, obesity and chronic diseases [GREEN PAPER]. Brussels, Belgium: COMMISSION OF THE EUROPEAN COMMUNITIES; 2005. 22 p. COM(2005), 637 final.
8. Helmrich SP, Ragland DR, Leung RW, Paffenbarger RS Jr. Physical activity and reduced occurrence of non-insulin-dependent diabetes mellitus. N Engl J Med
9. Hopkins WG. Measures of reliability in sports medicine and science. Sports Med
10. Le Faucheur A, Abraham P, Jaquinandi V, Bouyé P, Saumet JL, Noury-Desvaux B. Study of human outdoor walking
with a low-cost GPS
and simple spreadsheet analysis. Med Sci Sports Exerc
11. Nordbeck S. Computing distances in road nets. Pap Reg Sci Assoc
12. Paffenbarger RS Jr, Hyde RT, Wing AL, Hsieh CC. Physical activity, all-cause mortality, and longevity of college alumni. N Engl J Med
13. Rietveld P, Zwart B, van Wee B, van den Hoorn T. On the relationship between travel time and travel distance of commuters. Reported versus network travel data in the Netherlands. Ann Reg Sci
14. Saelensminde K. Cost-benefit analyses of walking
and cycling track networks taking into account insecurity, health effects and external costs of motorized traffic. Transp Res Part A Policy Pract
15. World Health Organization. A Physically Active Life through Everyday Transport with a Special Focus on Children and Older People and Examples and Approaches from Europe
. Copenhagen (Denmark): Regional Office for Europe, World Health Organization; 2002. p. 54.
16. World Health Organisation Regional Office of Europe Web site [Internet]. Copenhagen (Denmark): World Health Organisation Regional Office of Europe; [cited 2007 Nov 28]. Health Economic Assessment Tool for Cycling (HEAT for cycling). Available from: http://www.euro.who.int/transport/policy/20070503_1
17. Witlox F. Evaluating the reliability of reported distance data in urban travel behaviour analysis. J Transp Geogr