Analysis of Potential Determinants
of Cycle Commuting Speed
– With Special Reference to Gears, Showers, and
Ratings of Perceived Exertion
Erik Cunelius
THE SWEDISH SCHOOL OF SPORT
AND HEALTH SCIENCES
Master Degree Project 44:2020
Master Programme 2019-2021
Supervisor: Peter Schantz
Examinator: Magnus Lindwall
Abstract
Aim: The aim of this study is to analyse potential determinants of cycle commuting speed, with the following research questions: 1. How does cycle commuting speed relate to amount of gears, access to shower facilities, and rating of perceived exertion (RPE), when sex, age, weight, body mass index (BMI), duration, last digit in self-reported duration, and cycling area are controlled for? 2. How are the earlier found relationships between speed and duration, and speed and sex, affected when amount of gears, access to shower facilities, and RPE are added to the control variables age, weight, BMI, last digit in self-reported duration, and cycling area?
Method: In this cross-sectional study, 1526 adult cycle commuters (67% females) in Stockholm County, Sweden, were recruited through advertisements. In a
self-administered questionnaire, the respondents reported their sex, age, height, weight, duration of their typical cycle commuting journey, amount of gears, access to shower facilities at the destination, and rating of perceived exertion (measured by the Borg RPE Scale) while cycling. They also drew their cycling route on an individually adjusted map. Multiple linear regression was used to facilitate the analyses.
Results: In a regression model with the independent variables sex, age, weight, body mass index, duration, last digit in self-reported duration (1–4 or 6–9 compared to 0 or 5), cycling area (inner urban compared to suburban and suburban – inner urban), gears, shower facilities, and RPE, the dependent variable cycle commuting speed was
positively related to using a bicycle with five or more (as compared to four or fewer) gears, having convenient (as opposed to inconvenient or non-existent) access to shower facilities, and perceiving a higher degree of exertion while cycling. Gears, showers, and RPE were also found to substantially affect the earlier found relationships between speed and duration, and speed and sex. The respondents showed a clear tendency of giving odd-numbered, compared to even-numbered, ratings of perceived exertion. Conclusions: This study highlights the importance of gears, showers, and RPE in activities such as estimating, planning or taking measures in relation to cycle
commuting. It also shows that adding new factors to control for might change earlier established relationships within this field. When the Borg RPE Scale is used, correct instructions strengthens the validity.
Sammanfattning
Syfte och frågeställningar: Syftet med denna studie är att analysera potentiella påverkansfaktorer gällande cykelpendlingshastighet, med följande frågeställningar: 1. Hur relaterar hastighet till antal växlar, tillgång till duschfaciliteter och skattning av egenupplevd fysisk ansträngningsgrad (RPE), när kön, ålder, vikt, BMI, duration, slutsiffra i självrapporterad duration samt cykelområde kontrolleras? 2. Hur påverkas de tidigare funna förhållandena mellan hastighet och duration, samt hastighet och kön, när antal växlar, tillgång till duschfaciliteter och RPE adderas till kontrollvariablerna ålder, vikt, BMI, slutsiffra i självrapporterad duration, och cykelområde?
Metod: I denna tvärsnittsstudie rekryterades 1526 vuxna cykelpendlare (67% kvinnor) i Stockholms län genom annonser. I ett själv-administrerat formulär rapporterade de kön, ålder, längd, vikt, duration vid deras typiska cykelpendlingsresa, antal växlar, tillgång till duschfaciliteter vid destinationen, och skattning av egenupplevd fysisk ansträngningsgrad (enligt Borgskalan) under cyklingen. De ritade även in sin cykelväg på en individuellt anpassad karta. Multipel linjär regression användes för att
möjliggöra analyserna.
Resultat: I en regressionsmodell med de oberoende variablerna kön, ålder, vikt, BMI, duration, slutsiffra i självrapporterad duration (1–4 eller 6–9 jämfört med 0 eller 5), cykelområde (innerstad i jämförelse med ytterstad samt ytterstad – innerstad), växlar, duschmöjligheter och RPE, var den beroende variabeln cykelhastighet positivt
relaterad till användandet av en cykel med fem växlar eller fler (i jämförelse med fyra eller färre), att ha bekväm tillgång till dusch (i motsats till obekväm eller ingen tillgång alls) samt att uppleva en högre grad av fysisk ansträngning under cykling. Växlar, duschmöjligheter och RPE visade sig även substantiellt påverka de tidigare funna förhållandena mellan hastighet och duration, samt hastighet och kön. Respondenterna visade en tydlig tendens att ge ojämna, i jämförelse med jämna, RPE-skattningar. Slutsats: Denna studie belyser vikten av växlar, duschmöjligheter och egenupplevd ansträngningsgrad vid exempelvis bedömning, planering eller vidtagande av åtgärder i relation till cykelpendling. Den visar även att nya kontrollfaktorer kan förändra tidigare etablerade förhållanden inom detta fält. Korrekta instruktioner stärker validiteten vid användandet av Borgskalan.
Acknowledgements
First and foremost, I would like to express my deep and sincere gratitude to my research supervisor Prof. Peter Schantz, DrMedSc, research leader of The Research Unit for Movement, Health and Environment at The Swedish School of Sport and Health Sciences (GIH), whoseguidance and support has been nothing less than fantastic throughout the entire process. Thank you for your availability and dedication, friendly, patient and humble co-operation, wise advice, and the opportunity to build on your and your colleagues’ work. You are truly inspiring.
I would also like to recognize the highly valued contributions made by Ruggero Ceci, PhD, affiliated researcher at GIH, who has acted as a second supervisor, with the Borg rating scales as a particular area of expertise, Lasse ten Siethoff, PhD, GIH, for teaching me and giving input regarding statistics, course leader Maria Ekblom, DrMedSc, GIH, for writing advice, my friend Frida Magnusson for proofreading, and Sofia Cunelius, my wife, for invaluable efforts in favour of the well-being of the family. Finally, my special regards also goes to all the respondents who chose to participate in the research project. Thank you.
Table of contents 1 Introduction ... 1 2 Existing Research ... 3 2.1 Type of Bicycle ... 5 2.2 Access to Shower Facilities ... 6 2.3 Ratings of Perceived Exertion ... 6 2.4 Sex Differences ... 8 2.5 Concluding Remark ... 8 3 Aim and Research Questions ... 8 4 Materials and Methods ... 9 4.1 Study Design ... 9 4.2 Data Collection and Respondents ... 9 4.3 Characteristics of the Respondents ... 10 4.4 Ethics ... 12 4.5 Selection and Measurement of Variables ... 13 4.5.1 Distance or Duration ... 13 4.5.2 Duration ... 13 4.5.3 Distance and Speed ... 13 4.5.4 Geographical and Cycling Areas ... 14 4.6 Reliability and Validity ... 14 4.6.1 Duration, Distance, and Speed ... 14 4.6.2 Ratings of Perceived Exertion ... 15 4.7 Statistical Analyses ... 16 4.7.1 Assumptions ... 17 4.7.2 Detection of Highly Influential Data Points ... 18 4.7.3 Presentation ... 18 5 Results ... 19 5.1 Multiple Linear Regression Analysis With All Variables Included ... 19 5.2 Gears and Showers and Their Relationship With Cycle Commuting Speed ... 20 5.3 RPE and its Relationship With Cycle Commuting Speed ... 21 5.4 Effect on the Relationship Between Speed and Duration, and Speed and Sex ... 23
6 Discussion ... 25 6.1 Methodological considerations ... 28 6.2 Strengths and Limitations ... 29 6.3 Conclusions ... 29 References ... 31 Appendix A Litteratursökning ... 37 Appendix B PACS Q1 ... 38
1
1 Introduction
Cycling, and cycle commuting, is seen as a health enhancing, sustainable and environmentally friendly means of transport. There is strong evidence of its fitness benefits and it is also associated with improvements in relation to cardiovascular risk factors as well as with a lower all-cause and cancer mortality rate (Oja et al., 2011). Beside these aspects, cycle commuters have recurrently emphasized motives such as positive experiences, improved mood and stress relief (Hansen & Nielsen, 2014). The effects are prominent regarding pollution as well. In Stockholm County, a study of the potential has calculated that hundreds of years of life annually would be saved if all car commuters who live within a distance to work
corresponding to a bicycle ride no longer than 30 minutes would swap to this more active way of commuting (Johansson et al., 2017).
The advantages of cycling also seem to have elevated political interest in the phenomenon. In 2017, the Government of Sweden presented A national cycling strategy for more and safer cycling, the first of its kind (Ministry of Enterprise and Innovation, 2017). The aim was to encourage sustainable transport solutions, and contribute to improved public health, as well as to reduce both congestion and the environmental impact of travel (Ministry of Enterprise and Innovation, 2017). Clear incentives for a rise in cycling also exist regionally, and in the beginning of 2017, the number of counties in Sweden with an official cycling strategy had doubled compared to just two years before, which meant it had gone up to around 40% (Trafikverket, 2017). Stockholm County has a strategy (Trafikverket, 2014) in addition to an office dedicated to the promotion of cycling (Cykelkansliet), which was made permanent in 2019 (Region Stockholm, n.d.). A goal in the plan for Stockholm County is to markedly increase the amount of cycling so that the proportion of all journeys made this way will go from 5% to at least 20% by 2030 (Trafikverket, 2014; Statens institut för
kommunikationsanalys [SIKA], 2007).
Historically, the bicycle share of the total traffic has varied greatly throughout the years. In the 1940s, for example, cycling to work was a normality (Johnson, 2006; Schantz, 2018), and according to Emanuel (2012), during the first half of the 1940s, cycling accounted for a proportion of around 70%. This figure can be compared to around 20% in the early 1930s,
2
and under 1% in 1970. Much of the decline after World War II, Emanuel (2012) argues, could be ascribed to how cycling was treated politically.
Together with the goal of increased cycling, more insights are needed. In the Stockholm County cycle plan, commuting is identified as one of the key elements (Trafikverket, 2014), and in the national cycling strategy, extended knowledge about cycle commuting is called for (Ministry of Enterprise and Innovation, 2017). An increased understanding of cycle
commuting can be used in national economic, environmental, and health calculations, in urban and traffic planning, and to improve the way investments and interventions are carried out. More knowledge can also be beneficial when designing, for example, workplaces, by the consideration and layout of shared facilities such as bicycle storages and showers.
Furthermore, an extended understanding of cycle commuting can be used to make valuations of fitness and other health effects more accurate.
In the mapping of cycle commuting, just like in any means of transport, speed and velocity1 are important factors. If the displacement is given, which it is if someone commutes to the place of education or work, velocity will specify how long it will take to get there. Under normal circumstances, the speed will affect the velocity. For commuter cyclists, saving time has been identified as an important motive (Hansen & Nielsen, 2014). The speed (and in turn, the velocity) may therefore affect the tendency to choose cycling over other means of transport (Heinen, van Wee, & Maat, 2010). Thus, increased knowledge about factors that influence cycling speed could facilitate improvements of the conditions for cycle commuting and make it more appealing.
1 Although used interchangeably by some, in this thesis (when not referring to work done by others), ‘speed’ and ‘velocity’ are going to be used like in physics, by definitions formulated by Jones (2020): Speed is the distance travelled by unit of time, while velocity is the
displacement by unit of time. Speed is a scalar quantity and velocity a vector quantity. An example: “If ran across the room and then returned to your original position, you would have a speed — the distance divided by the time. But your velocity would be zero since your position didn't change between the beginning and the end of the interval. There was no displacement seen at the end of the time period” (Jones, 2020, Speed vs Velocity).
3
2 Existing Research
In a review of the science of cycling, successful elite cyclists were found to generally have a high maximal aerobic power output combined with an ability to work at relatively high power outputs for extended periods of time (Atkinson, Davison, Jeukendrup & Passfield, 2003; Faria, Parker & Faria, 2005). However, regarding race velocity, the factors shown in Figure 1 were also highlighted as important:
Figure 1. ”Diagram showing the various factors that influence cycling power output and speed” (Atkinson et al., 2003, Figure 1)
Bike design and race-specific retarding factors (e.g. hills and wind) influence the velocity directly (Atkinson et al., 2003). On the other hand, some of the factors exert their effect indirectly, by the influence on cycling power, which in turn is central for the velocity.
The model above is put together with competing elite cyclists in mind. However, it is likely that commuters cycle under conditions that are different in a number of decisive ways. For example, highest possible average velocity is presumably not their one (and only) priority. What science is there then about the non-competitive type of cycling, or cycling with the purpose of commuting?
Overall race velocity
Bike design Pacing strategy Race-specific inherent physiological ability Race-specific nutritional strate res Race-specific training strate res Cycling power Rider position Race-specific retarding forces
4
Non-competitive cycling speed has been found to vary substantially when measured in different ways and in different places and settings (El-Geneidy, Krizek & Iacono, 2007). Higher average speeds have been recorded on cycle paths that are wider (above 3.5m), have a centreline, where there is a visual segregation between cyclists and pedestrians (Boufous, Hatfield, & Grzebieta, 2018), and where the cycling paths are separated from the street (El-Geneidy et al., 2007).
In a Swedish study, Eriksson, Forsman, Niska, Gustafsson and Sörensen (2019) found the average speed of non-competitive cyclists to vary depending on the formation and gradient of the road, as well as the composition of types of road users. Not surprising, lower speeds were found going uphill, near intersections, and on paths with a high flow of pedestrians. Higher speeds were found going downhill or on commuter routes. These findings confirm the ones made by Manum, Nordström, Gil, Nilsson and Marcus (2017), who report lower median speeds to correlate with signal crossings, multiple entrances along the segment, mixed use of the cycling path (both cyclists and walkers), and sharper turns. Higher speeds were found to correlate with downhill slopes, two-way bicycle lanes, and longer segments where braking was mostly not needed.
As suggested in Figure 1, not only external factors may influence the speed, but even
conditions relating to the cyclist. Examining commuters, Schantz (2017) found the following: “Cycling speeds were positively related to commuting distances or durations, being male, of younger age, having higher body weight but lower body mass index (BMI), and using the last digits 1–4 or 6–9 in duration reports (as compared to 0 and 5), as well as cycling in suburban (versus inner urban) areas.” (Schantz, 2017, p. 1.) These findings are in line with the ones made by Boufous et al. (2018) regarding sex and age and with the results brought forward by El-Geneidy et al. (2007) regarding distance.
The positive relationship between average speed and longer distances or durations is something Schantz, Salier-Eriksson and Rosdahl (in press; P. Schantz, personal
communication, 13 December 2019) reflects upon. Based on the relationship between cycling speed and oxygen consumption (Atkinsson et al. 2003; Faria et al. 2005), Schantz and co-workers put forward three possible explanations: Firstly, that the exercise intensity relative to maximal oxygen consumption increases with growing cycling distance or duration. Secondly, that people who cycle longer have higher levels of maximal oxygen uptake. Thirdly, a
5
combination of both these explanations (i.e. that both the relative intensity as well as the levels of maximal oxygen uptake are higher in commuters who cycle longer). They also write that this deserves to be examined in future studies, because it at present leaves us with
uncertainty regarding the relative intensity category and the levels of maximal oxygen uptake for different cyclist groups. One way of addressing the uncertainty about the relative intensity could be to look into the degree of perceived exertion (Robertson & Noble, 1997). Therefore, the relationship between perceived exertion, measured by the Borg RPE (Ratings of Perceived Exertion) Scale (Borg, 1970), and average speeds for cyclists commuting with varying
durations, is something I intend to examine in this study.
Also Hansen and Nielsen (2014) stress the need of a more thorough understanding of factors governing long distance commuter cycling. They have examined the differences in
characteristics of the cyclists and report that people who commute longer than 5 km (i.e. longer than the 75th percentile in this particular, Danish setting) systematically have “more options/choices for mobility (parking access and driving licences), higher incomes and higher educations.” (Hansen & Nielsen, 2014, p. 63) Could these characteristics influence speed somehow?
2.1 Type of Bicycle
As presented above, speed is thought to be influenced by bike design. However, such a relationship could not, due to methodological reasons, be determined by Eriksson et al. (2019). Also,the relationship between type of bicycle and speed claim among commuters remains somewhat uncertain (Eriksson, Niska, Sörensen, Gustafsson & Forsman, 2017; Kircher, Nygårds, Ihlström, & Ahlström, 2017). Furthermore, Eriksson et al. (2017) find it probable that the type of bicycle is connected with private economy. Therefore, given that cyclists who commute longer have higher incomes (Hansen & Nielsen, 2014), one of the possible explanations for their higher average speeds (Schantz, 2017; El-Geneidy et al., 2007), could be a more frequent use of more efficient (and more expensive) bicycle types. Also, the more time people spend on their bicycles, the more willing they might be to invest financially in them. One particular aspect of the bike design that presumably has an effect on speed is the amount of gears. The relationship between cycle commuting speed and amount of
6
gears, while controlling for a number of other factors including duration, will be investigated in this study.
Compared to users of classic bicycles, Schleinitz, Petzoldt, Franke-Bartholdt, Krems, and Gehlert (2017) found that people on electrically assisted bicycles (electric bicycles/e-bikes) generally travel faster. For reasons stated in the Data Collection and Respondents section, only fully active commuters will be included in this study.
2.2 Access to Shower Facilities
What is the relationship between speed and the presence of shower facilities at the end of the commute? Perhaps people allow themselves to exercise at a higher intensity and cycle faster knowing there is an opportunity to shower immediately afterwards? In an overview of the literature, Heinen et al. (2010) found the influence of having shower facilities at work on the likelihood of cycle commuting to be ambiguous. While some studies have found it important (e.g. Hunt & Abraham, 2007), others have not (Taylor & Mahmassani, 1996; Stinson & Bhat, 2004). Previous research has focused more on the influence of access to shower facilities on the choice of means of transport, and less on the correlation between prevalence of such facilities and speed. This study intends to examine the latter.
2.3 Ratings of Perceived Exertion
The degree of exertion can be measured in several ways, one of the most used being the Borg RPE (Ratings of Perceived Exertion) Scale. This is because our own perceived exertion has been proved to measure exercise intensity very well (Borg, 1998). When cycling at a constant workload, the ratings of perceived exertion have been found to become higher as the duration of the exercise lengthens (Ljunggren, Ceci, & Karlsson, 1987; Kang et al. 1996). This implies the importance of taking duration into account when evaluating RPE in relation to commuter cyclists’ journeys. In the study by Ljunggren et al. (1987), the participants exercised at a power level inducing a 4 mmol/l blood lactate concentration (WOBLA), and in the study by Kang et at. (1996), the cyclist performed at 70% of their peak oxygen uptake (VO2peak) until exhaustion.
7
Eston (2012) also points out the value of duration in relation to RPE, and highlights the ratings as an inner way to regulate pace:
Research also shows that the rate of increase in RPE during self-paced competitive events of varying distance, or constant-load tasks where the participant exercises until volitional exhaustion, is proportional to the duration that remains. These findings suggest that the brain regulates RPE and performance in an anticipatory manner based on awareness of metabolic reserves at the start of an event and certainty of the
anticipated end point. Changes in pace may be explained by a continuous internal negotiation of momentary RPE compared with a preplanned “ideal rate of RPE progression” template, which takes into account the portion of distance covered and the anticipated end point. (Eston, 2012, p. 175)
Theoretically, this would allow higher average speed at shorter distances, in contradiction to the findings of Schantz (2017) and El-Geneidy et al. (2007) regarding commuter cyclists. However, as expressed earlier: commuter cyclists do not necessarily pace themselves to reach the highest average speed possible, but other factors are likely to also play a role.
The influence of race-specific nutritional strategies on performance, put forward by Atkinson et al. (2003) above, is confirmed by the findings of Kang et al. (1996), who found the ratings of perceived exertion (measured by the Borg RPE Scale) to differ depending on the
availability of carbohydrate substrate for cyclists performing at (as mentioned earlier) 70% of their peak oxygen uptake (VO2peak) until exhaustion. However, no statistically significant differences were found for the first 80 minutes of exercise. Consequently, carbohydrate supplementation are not expected to exert much of an influence on the ratings of perceived exertion, and therefore not even on the speed, for commuters cycling for 50 minutes or less (which will be the span of durations used in this study). In comparison, average single mode (as opposed to double mode, i.e. commuters that sometimes cycle and sometimes walk, who have been identified as a different subgroup, see Stigell and Schantz [2015]) cycle commuters (with regard to mean age and route distances for females and males respectively) in Schantz’s study from 2017, when tested later, cycled at a relative exercise intensity of about 65% of maximal oxygen uptake (Schantz et al., in press).
8
2.4 Sex Differences
In order to explain the variation in commuters’ cycling speed, sex2 stands out as a vital factor to take into account. This is because Schantz (2017) found it to be decisive, and that general differences between men and women have been registered regarding levels of maximum oxygen uptake. Other reasons are that a greater proportion of men, compared to women, have been reported to use shower facilities at work (70% compared to 58%), special clothes for cycling (71% compared to 55%), and a specialized bicycle (e.g. road racer or cyclo-cross bikes compared to standard bikes) when cycling to work (34% compared to 15%) (Hansen & Nielsen, 2014). Hansen and Nielsen (2014) argue that the women’s choice of bicycles
generally make it less possible to attain higher speeds, which may to a certain degree, in addition to other factors, explain why they, on average, do not cycle as fast as men.
2.5 Concluding Remark
This review of the existing research in the field of cycle commuting reveals some gaps of knowledge regarding a number of certain aspects. By including average speed and type of bicycle (in the form of amount of gears), access to shower facilities, and rating of perceived exertion in the same regression model, while controlling for sex, duration and other variables already shown to be important, this study intends to shed light on areas that have not been investigated before.
3 Aim and Research Questions
The aim of this study is to analyse the relationships between cycle commuting speed (speed) and type of bicycle (gears), access to shower facilities (showers), and rating of perceived exertion (RPE), as well as to investigate how these relationships might affect the earlier found relationships between speed and duration, and speed and sex, when other previously noted determinants are controlled for. The research questions are:
2 In this piece of work, no distinction is made between ’sex’ and ‘gender’. Instead, ‘sex’ is used as an overriding term including both biological and socio-cultural factors associated with being male or female, man or woman.
9
1. How does speed relate to gears, showers, and RPE, when sex, age, weight, body mass index (BMI), duration, last digit in self-reported duration, and cycling area are controlled for?
2. How are the earlier found relationships between speed and duration, and speed and sex, affected when gears, showers, and RPE are added to the control variables age, weight, BMI, last digit in self-reported duration, and cycling area?
4 Materials and Methods
4.1 Study Design
In this cross-sectional study, multiple linear regression will be used to facilitate analyses of the relationship between average cycle commuting speeds, hereafter also referred to as speed, and the variables type of bicycle (gears), access to shower
facilities (showers) and rating of perceived exertion (RPE, measured by the Borg RPE Scale). The model will also include all the statistically significant variables in a model built by Schantz (2017): sex, age, weight, body mass index (BMI), duration, last digit in self-reported duration (1–4 or 6–9 compared to 0 or 5), and cycling area (inner urban compared to suburban and suburban – inner urban), to account for potential
confounders. Multiple linear regression, and more specifically: an analysis of the change of the unstandardized coefficients (B), will also be used as a method to answer the second research question.
4.2 Data Collection and Respondents
At the Swedish School of Sport and Health Sciences (GIH) in Stockholm, Sweden, a multidisciplinary research project with the title Physically Active Commuting in Greater Stockholm (PACS, https://www.gih.se/pacs) has been running for a number of years. The ground data material in this study is retrieved from a paper-based questionnaire (PACS Q1) that was created and sent out within this framework. PACS Q1 is self-administered, in Swedish, and contains 35 question items, of which some are used in this study. (A version of PACS Q1 translated into English is found in Appendix B.) Schantz provides information
10
about the data collection process for his study published in 2017 (on which the current study is building on): The recruitment of respondents began in 2004 through advertisements in two large newspapers in Stockholm (Dagens Nyheter and Svenska Dagbladet). There was no provision of incentives. Inclusion criteria were: “a minimum age of 20 years, residency in the County of Stockholm (except for the municipality of Norrtälje), and to, at least once a year, cycle or walk the whole way to one’s place of work or study.” (Schantz, 2017, p. 3.) This resulted in an original number of 2148 responses to the advertisement, received between September and December that same year. The response rate was 93,8%, as in 133 cases, the questionnaire was never returned. Another 21 of the responders did not meet the inclusion criteria, and 276 were not included because they were single-mode pedestrians. 56 people (my count) were excluded because of missing values in at least one of the studied variables. This left 1662 (my count) responders for Schantz’s (2017) study that hence met the inclusion criteria, were commuting by bicycle and filled in and returned the questionnaire in a correct way. However, because the current study includes a partly different set of variables, the total number of respondents this time is 1526. The data file with all the data from the questionnaire was sent to me in electronic form from P. Schantz in January 2020. In the study by Schantz, only fully active cycling was analysed. Consequently, users of electric bicycles were
excluded. The same delimitation will therefore apply this time.
4.3 Characteristics of the Respondents
Characteristics of the respondents with regard to cycling speed, age, height, weight, body mass index (BMI), duration of the cycle journey from home to the place of education or work, and rating of perceived exertion (RPE), for female and male commuters, are shown in Table 1. Table 2, 3, 4, and 5 display how the respondents are categorized in relation to last digit in self-reported duration, cycling area, type of bicycle (gears), and access to shower facilities (showers) respectively. The reason behind presenting the statistics of females and males separately is the identification of sex as an influential factor in explaining the variation in cycling speed.
11
Table 1. Characteristics of the respondents, reported for females and males, with mean, standard deviation, and minimum and maximum values (total n = 1526)
Sex Mean Std. Deviation Minimum Maximum
Female (n = 1030) Cycling speed (km/h) 14.1 3.55 3.75 29.4
Age (years) 46.6 10.8 20 73
Height (cm) 168 6.11 148 187
Weight (kg) 64.6 8.61 44 112
BMI (kg/m2) 22.9 2.76 16.2 36.8
Duration (min) 23.0 11.6 2 50
RPE (Borg RPE Scale) 12.0 2.20 6 18
Male (n = 496) Cycling speed (km/h) 17.7 4.49 6.23 31.7
Age (years) 47.0 11.1 21 73
Height (cm) 181 6.48 163 198
Weight (kg) 78.3 8.79 56 105
BMI (kg/m2) 24.0 2.38 18.4 33.7
Duration (min) 26.1 11.8 1 50
RPE (Borg RPE Scale) 13.0 2.12 6 19
Table 2. Categorization of the respondents in relation to last digit in self-reported duration (total n = 1526)
Sex Frequency Percent
Female (n = 1030) Last digit 1–4 or 6–9 268 26.0 Last digit 0 or 5 762 74.0
Total 1030 100
Male (n = 496) Last digit 1–4 or 6–9 161 32.5 Last digit 0 or 5 335 67.5
Total 496 100
Table 3. Categorization of the respondents in relation to cycling area (total n = 1526)
Sex Frequency Percent
Female (n = 1030) Suburban and suburban – inner urban 800 77.7
Inner urban 230 22.3
Total 1030 100
Male (n = 496) Suburban and suburban – inner urban 408 82.3
Inner urban 88 17.7
12
Table 4. Categorization of the respondents in relation to type of bicycle (gears) (total n = 1526)
Sex Frequency Percent
Female (n = 1030) Bicycle with 0-4 gears 425 41.3 Bicycle with five gears or more 605 58.7
Total 1030 100
Male (n = 496) Bicycle with 0-4 gears 86 17.3
Bicycle with five gears or more 410 82.7
Total 496 100
Table 5. Categorization of the respondents in relation to access to shower facilities (showers) (total n = 1526)
Sex Frequency Percent
Female (n = 1030) Inconvenient access or no access at all 353 34.3
Convenient access 677 65.7
Total 1030 100
Male (n = 496) Inconvenient access or no access at all 126 25.4
Convenient access 370 74.6
Total 496 100
4.4 Ethics
As part of the larger research project, this study has been approved by the Ethics Committee North of the Karolinska Institute at the Karolinska Hospital (Dnr 03-637), Stockholm, Sweden, and the respondents have given their written informed consent. The ethical aspects of this particular study have been thoroughly assessed separately, and no potential, significant drawbacks for the participants, others or the society as a whole have been detected. No
individual can be identified. The study does not require any additional effort or information from the participants, but is a way to extract as much relevant knowledge as possible from the data already provided. The underlying aim is to deepen the understanding of cycle
commuting, which by extension also could lead to improved conditions for cycling. In that way all cycle commuters, as well as the environment and society (see the introduction), could possibly benefit from the outcomes.
13
4.5 Selection and Measurement of Variables
The selection of variables for this study is made from the items in the PACS Q1
questionnaire. Sprung from existing research, the variables chosen have either been found to be influential, or are assessed to have a particular potential to have a significant effect. Below, some of the decisions are portrayed even further.
4.5.1 Distance or Duration
The research of Schantz (2017) generated two multiple linear regression models with cycling speed as the dependent variable. Both models have an identical set of independent variables, with the exception of distance (model 1) and duration (model 2) respectively. Because duration in previous research has been detected as a vital factor to include in relation to RPE, this study will proceed from model 2.
4.5.2 Duration
The durations in this study have been obtained from self-reports. The respondents were asked to record and fill in the duration (in hours and minutes) of their cycle commuting travel on a normal day when no errands were undertaken on the way. Although many cyclists commute both to and from the place of education or work, only the duration reports of the way there are included in this study. This is partly for the sake of conformity with the study by Schantz (2017), and partly because of the involvement of access to shower facilities away from home as a variable.
4.5.3 Distance and Speed
Participants first drew their commuting route on an individually adjusted map. Then, the distance was measured with a digital curvimetric instrument (Run Mate Club, CST/Berger, Watseka, IL, USA) two or more consecutive times by a technical assistant, in order to make it accurate (Schantz, 2017). For further descriptions of the method and the procedure, see Schantz (2017), and Schantz and Stigell (2009). The average speed for each respondent has been calculated by dividing the person’s commuting route distance by the self-reported duration.
14 4.5.4 Geographical and Cycling Areas
All participants were residing and commuting within the Stockholm County, Sweden, with the exclusion of the Municipality of Norrtälje. By the time of the data collection (2004), this specified region had around 1.8 million inhabitants (Eurostat, 2020; Statistics Sweden [SCB], 2014). The cycling area dummy variable of inner urban versus suburban and suburban – inner urban commuters, present in Schantz’s (2017) study, will be used once again. However, the variable of suburban – inner urban (i.e. commuters living in one of these areas and studying or working in the other) versus inner urban and suburban commuters will not, because it was not statistically significant at the p ≤0.05 level in the multiple linear regression analysis. The boundary between the cycling areas, as well as a more extensive description of the entire region, including topography, urban formation and residential density, is found in Schantz (2017), Wahlgren, Stigell, and Schantz (2010) and Wahlgren and Schantz (2011). Where each participant was commuting in relation to the different cycling areas was based on the postal area codes of his or her home address and place of education or work.
4.6 Reliability and Validity
4.6.1 Duration, Distance, and Speed
The fact that the durations in this study have been obtained from self-reports limits the validity. In the same data material as this study is based on, Schantz (2017) noted a tendency of the commuter cyclists’ travel times to be rounded off so that the last digit values of 0 and 5 where much more frequent compared to 1–4 and 6–9. Hence, self-reported travel times of e.g. 25, 30, 35 etc. minutes were much more common than e.g. 24, 28 or 36. This is in line with previous research regarding self-reporting of travel times in general (Rietveld, 2002; Kelly 2013). Self-reports of duration have also been found to generally be too long (Kelly, Krenn, Titze, Stopher, & Forster, 2013). Moreover, Kelly (2013) states the systematic bias of over-reporting to be even higher when the travel time is rounded to an integer with last digit 0 or 5, and when the journey duration is longer. Schantz (2017) found males to report the travel time duration with the last digit of 1–4 or 6–9 to a greater extent, and that this difference between the sexes become even more evident with durations that were increasing. In summary, this section brings reasons why true durations might not be exactly determined by self-reports. However, the durations stated in the PACS Q1 questionnaire are judged to be of satisfactory use in the fulfilment of the aim of this study.
15
Considering how speed has been calculated, with duration as an important part, limitations to the validity will apply also regarding this variable. The criterion method for measuring route distances, however, has been found to have a high validity and reproducibility (Schantz & Stigell, 2009), even though it does not consider the addition of distance due to differences in elevation (P. Schantz, personal communication, 21 April 2020).
In this study, speed and duration are combined in the same regression model. When doing so, it is important to be aware of their definitional relation, in order to interpret the results in a valid way. The relationship between speed (s), distance (d) and time/duration (t) can be expressed by the formula s = d/t. Accordingly, some influence of duration on speed is set by definition. If the distance is given (which it normally is when someone commutes to the place of education or work), and all other factors are kept constant, the formula makes it evident that longer duration inevitably will result in lower average speed. In reality, not all other factors are kept constant, and how variations in some of these factors correlate with speed is what this study will examine. The fact that Schantz (2017) found that average cycling speeds were increasing (as opposed to decreasing) with longer durations makes it clear that other factors do have an effect, thus eliminating any doubt that the relationship between speed and duration can entirely be ascribed to their inherent dependence.
4.6.2 Ratings of Perceived Exertion
The idea behind the development of the Borg RPE Scale was “to enable reliable and valid estimations of perceived exertion” (Borg, 1998, p. 13) and Borg himself claims that countless studies have shown it to function very well (Borg, 1998).
Regarding the validity of the scale to assess inter-individual differences in perceived exertion, Robertson and Noble (1997) write:
The scaling rationale . . . holds that as exercise intensity increases from very low to maximal levels, there is a corresponding and equal increase in perceptual intensity from “no exertion” to “maximal exertion”. In this case, any two clinically normal individuals can respond with a perceptual rating that falls at 50% of their perceptual response range, even though the absolute power output (i.e. exercise stimulus) associated with that perceptual rating is higher for the individual with the
16
responses between individuals differing in level of maximal aerobic power is, therefore, possible. (Robertson & Noble, 1997, p. 412)
Thus, to compare perceived exertion responses of people of, for example, different aerobic capacities, sexes, ages, weights, and body mass indexes, is perfectly doable. It is such comparisons this study intends to do.
Borg emphasises that the ratings may be affected by factors such as personality, motivation, and over- and underestimation tendencies, although factors related to the exercise intensity itself, like the signals from muscles and chest, are what matters the most (Borg, n.d.). The ratings have also been found to vary depending on other types of stimuli, such as the
environment in which the exercise takes place. Average running and cycling speeds have been found to be higher outdoors compared to in a laboratory environment indoors, when
exercising at a pre-determined level of the Borg RPE Scale (Ceci & Hassmén, 1991; Mieras, Heesch, & Slivka, 2014). These results stress the reason to compare RPE derived from similar settings. In this study, all respondents obviously cycle outdoors.
In order for the RPE Scale to function in an optimal way, Borg urges the importance of correct instructions (Borg, 1998; Borg, n.d.). The limited instructions to the respondents in this study (see question 11 in PACS Q1, Appendix B) might therefore reduce the validity, which is something to keep in mind when interpreting the results. In this study, the original Borg RPE Scale from 1970 is used (Borg, 1998).
In order to strengthen the validity of the statistical analyses, great emphasis has been put on the careful inspection of all data, the choice of appropriate statistical tests, and the assurance that all assumptions of each test are met, as described below.
4.7 Statistical Analyses
The software package IBM SPSS Statistics, version 26, was used to process and analyse the data. A frequency distribution was run for each of the included variables to check for potential errors.
17 4.7.1 Assumptions
Prior to analysis, examinations were made to ensure that the assumptions for multiple linear regression were met. Boslaugh (2012) lists the following assumptions, which will be
discussed in more detail below: data appropriateness, linearity, independence, distribution, homoscedasticity, independence and normality of the errors, and multicollinearity.
In order to meet the assumption of data appropriateness, both the ordinal variable gears and the nominal variable showers were recoded. This was because both of them, in their original form, had more than two categories and that the ordinal variable was far from equally spaced (for the possible answers regarding these two items in PACS Q1, see questions 6 and 17 in Appendix B). For valid interpretations, variables with such conditions should not be entered in a linear regression analysis. Regarding gears, the results of a one-way ANOVA, including homogeneity of variance, Welch, and Games-Howell (post hoc) tests, showed a statistically significant (p < .001) difference between the average speed of commuters using bicycles with five or more gears compared to the ones using bicycles with fewer, but not between
commuters using bicycles with no gears at all and those using bicycles with two to four gears (p = .331). Consequently, a dichotomous variable was created with between four and five gears as the cutting point. The reason for conducting the Welch and Games-Howell tests was that the one-way ANOVA assumption of homogeneity of variance was not considered to be satisfied (p < .001).
The one-way ANOVA assumption of homogeneity of variance was considered to be violated (p < .001) also for the variable showers. This time, the Welch and Games-Howell tests revealed a statistically significant (p < .001) difference between the average speed of one group (namely commuters with convenient access to shower facilities), and each of the other two (commuters with either no access to shower facilities or access that was reported as inconvenient). There was no statistically significant difference between the latter two groups themselves (p = .607). A dichotomous variable was therefore created, where the two groups with no significant difference between them in terms of average speed were merged into one.
The assumption of linearity was checked by graphing the relationship between the dependent and the independent variables in scatter plots with added fit and interpolation lines. As found by Schantz (2017), the positive correlation between cycling speed and duration appears to be
18
linear from low levels up to a certain section, with an altered relationship afterwards.
Therefore, in Schantz’s (2017) study, only durations ≤ 50 minutes were used. For the sake of conformity, and to avoid violating the assumption of linearity, the same decision will apply in this study.
The rest of the listed assumptions were considered to be met after having completed the following examinations: Independence of each value of the dependent variable was
considered upheld based on knowledge of the participants, the data and how it was collected. The distributions of the data were checked in frequency tables and histograms. A residual plot revealed no problems with homoscedasticity, the Durbin-Watson statistic (d = 1.94) made the same regarding independence of the errors, and a predicted probability (P-P) plot visualised an approximate normality of the errors. The variance inflation factor (VIF) indicated
acceptable levels of multicollinearity. All values were ≤ 4.72, with a mean of 1.88 (my calculation).
With all the assumptions checked, and necessary adjustments made, conditions for multiple linear regression analyses were found present. All predictors were entered in the model simultaneously (Forced entry/Enter method).
4.7.2 Detection of Highly Influential Data Points
Cook’s distance (all values < 0.02; mean < 0.01) was used to detect data points with particularly high influence on the regression line, with the most influential points being further inspected. The comparatively low maximal value, in combination with the judgement that no point was too far away from another, led to the decision to keep the data set intact.
4.7.3 Presentation
For the main regression model, the presented results include,: R2 and R2adjusted, unstandardized coefficients (B), 95% confidence intervals (CI) for B, p-values, and variance inflation factor (VIF). The level of significance (α) was set to the common standard of 5% (p ≤ 0.05). The standardized coefficients (β) as well as zero-order, partial and part correlations are not presented, due to the either dichotomous or ordinal nature of the variables particularly under investigation. Regarding the multi-levelled variable RPE, also mean cycling speeds are
19
shown, together with standard deviation, standard error of the mean, 95% confidence interval for mean (lower and upper bounds), as well as minimum and maximum values.
5 Results
The results are based on the response from a total of 1526 adult cycle commuters in Stockholm County.
5.1 Multiple Linear Regression Analysis With All Variables Included
A multiple linear regression analysis of the dependent variable cycle commuting speed and the independent variables sex, age, weight, body mass index (BMI), duration, last digit in self-reported duration (1–4 or 6–9 compared to 0 or 5), cycling area (inner urban compared to suburban and suburban – inner urban), gears, showers, and RPE are seen in Table 6. All variables were entered in the model at the same time (Forced entry/Enter method). Presented statistics include: unstandardized coefficients (B), standard error, level of significance (Sig.), 95% confidence interval for B (lower bound and upper bound), and variance inflation factor (VIF).The mean speed for all respondents was 15.2 km/h. A significant regression equation (F(10,1515) = 104, p = < .001) was found, with an R2 of .41 (R2adjusted= .40). Respondents’ predicted cycle commuting speed is equal to 13.3 + 2.13 (sex) – 0.065 (age) + 0.049 (weight) – 0.185 (BMI) + 0.058 (duration) – 1.39 (last digit in self-reported duration) – 1.43 (cycling area) + 1.55 (gears) + 1.31 (showers) + 0.263 (RPE), where sex is coded as 0 = Female, 1 = Male, age is measured in years, BMI is calculated by kg/m2, duration is measured in minutes, last digit in self-reported duration is coded 0 = 1–4 or 6–9, 1 = 0 or 5, cycling area is coded 0 = Suburban and suburban – inner urban, 1 = Inner urban, gears is coded 0 = zero to four gears, 1 = Five gears or more, showers is coded 0 = No or inconvenient, 1 = Convenient, and RPE is measured by the Borg RPE Scale (6–20). All independent variables in the model were statistically significant (p = <.05) predictors of cycle commuting speed.
20
Table 6. Multiple linear regression analysis of the relationship between the dependent variable speed and ten predictor variables, for cases with a self-reported duration of ≤ 50 minutes (total n = 1526) Predictor Variables Unstandardized Coefficients Sig. 95,0% Confidence Interval for B Collinearity Statistics B Std. Error Lower Bound Upper Bound VIF (Constant) 13.3 .987 .000 11.3 15.2
Sex (0 = Female; 1 = Male) 2.13 .264 .000 1.61 2.65 2.18
Age (years) -.065 .008 .000 -.080 -.049 1.10
Weight (kg) .049 .017 .004 .016 .082 4.72
BMI (kg/m2) -.185 .056 .001 -.294 -.075 3.23
Duration (min) .058 .009 .000 .040 .075 1.59
Last digit in self-reported duration (0 = 1–4 or 6–9; 1 = 0 or 5)
-1.39 .197 .000 -1.78 -1.01 1.12
Cycling area
(0 = Suburban and suburban - inner urban; 1 = Inner urban)
-1.43 .221 .000 -1.86 -.991 1.15
Gears
(0 = 0–4 gears; 1 = Five gears or more)
1.55 .190 .000 1.18 1.92 1.14
Showers (0 = No or inconvenient access; 1 = Convenient access)
1.31 .186 .000 .948 1.68 1.06
RPE (Borg RPE Scale) .263 .046 .000 .173 .354 1.49
5.2 Gears and Showers and Their Relationship With Cycle
Commuting Speed
A multiple linear regression analysis, including the variables seen in Table 6, revealed gears to be a statistically significant (p < .001) predictor of cycle commuting speed. In the model, using a bicycle with five or more gears increased the commuter’s expected speed with 1.55 km/h (unstandardized coefficient [B]), compared to using a bicycle with fewer gears.
In the same model, the variable showers was also shown to be a statistically significant (p < .001) predictor of speed. Having reported convenient access to shower facilities at the place of education or work was tied with an increased expected cycling speed of 1.31 km/h
21
(unstandardized coefficient [B]), compared to having stated the access to be less convenient or non-existent.
5.3 RPE and its Relationship With Cycle Commuting Speed
The mean cycling speeds for commuters, together with standard deviation, standard error of the mean, 95% confidence interval for mean (lower and upper bounds), as well as minimum and maximum values, grouped according to their given rating of perceived exertion
(measured by the Borg RPE Scale), are seen in Table 7. The Borg RPE Scale ranges from 6 to 20 (Borg, 1998).
Table 7. Cycling speed for commuters with different ratings of perceived exertion, for cases with a self-reported duration of ≤ 50 minutes (total n = 1526)
RPE (Borg RPE Scale) n Mean Std. Deviation Std. Error
95% Confidence Interval for Mean
Minimum Maximum Lower Bound Upper Bound
6 14 11.5 3.54 .947 9.43 13.5 6.00 16.0 7 60 13.7 4.62 .596 12.5 14.9 6.23 26.5 8 20 13.4 4.03 .902 11.5 15.3 6.00 21.8 9 111 12.3 3.54 .336 11.6 12.9 4.16 21.6 10 29 14.0 3.09 .574 12.8 15.2 6.45 20.5 11 254 14.1 3.66 .230 13.7 14.6 3.75 27.6 12 150 14.8 3.43 .280 14.2 15.3 4.44 25.8 13 524 15.0 3.60 .157 14.7 15.3 5.81 29.4 14 137 16.9 4.36 .373 16.2 17.7 4.80 31.7 15 150 18.3 4.19 .342 17.6 18.9 4.05 31.7 16 52 19.0 4.72 .655 17.7 20.3 6.68 28.7 17 19 21.4 4.46 1.02 19.3 23.6 12.6 28.1 18 5 22.1 4.09 1.83 17.0 27.1 16.6 26.4 19 1 24.2 . . . . 24.2 24.2 Total 1526 15.2 4.24 .109 15.0 15.5 3.75 31.7
As seen in Table 7, from the rating of 9 and onwards, the mean cycling speed of the
commuters is steadily increasing with every rising step on the rating scale. In a multiple linear regression analysis, also including the variables shown in Table 6, RPE was found to be a statistically significant (p < .001) predictor of cycle commuting speed. In the model, every
22
one unit increase on the RPE scale corresponds with an expected increase of the cycling speed by 0.263 km/h (unstandardized coefficient [B]). Nearly the entire range of the scale (6-20) came to use (6-19).
Noteworthy in Table 7 is also that the number of respondents giving a rating with an odd number seems to be clearly higher compared to the frequency of the ratings that are even. To examine how likely it is that this distribution was due to chance, a chi-square goodness-of-fit test was performed. With all even-numbered ratings (6, 8, 10, 12, 14, 16, 18, 20) put together in one group and all odd-numbered ratings (7, 9, 11, 13, 15, 17, 19) in another, a new variable was created. A comparison between the expected (equal, although there is one more even-numbered level in the scale) and detected proportions of these two groups revealed a disproportionately high frequency of ratings with an odd number (p < .001), as shown in Table 8. The test was statistically significant: χ2(1) = 332, p < .0001. Therefore, we could reject the null hypothesis (assuming no significant difference between expected and detected values) and conclude that, in this sample, there is a statistically significant difference in the preference of giving an odd-numbered (n = 1119) compared to an even-numbered (n = 407) rating. Phi (φ) was used as a measure of effect size, and φ = .47 (my calculation) could be
interpreted (with some caution) as being moderate, on the verge of high (Boslaug, 2012; Statistics Solutions, 2020).
Table 8. Chi-square goodness-of-fit test with observed and expected distributions of even-numbered versus odd-even-numbered ratings of perceived exertion for cases with a self-reported duration of ≤ 50 minutes (total n = 1526)
Observed n Expected n Residual
RPE with an even number 407 763 -356
RPE with an odd number 1119 763 356
23
5.4 Effect on the Relationship Between Speed and Duration, and
Speed and Sex
In a regression model with the dependent variable speed and the independent variables sex, age, weight, BMI, duration, last digit in self-reported duration, and cycling area (i.e. all the statistically significant variables in Schantz’s model 2 from 2017), here referred to as the original model, the unstandardized coefficient (B) for each of these predictors decreased when gears, showers, and RPE were added at the same time, forming what here is referred to as the final model. In other models, gears, showers, and RPE were entered in one model each. Table 9 displays how the unstandardized coefficients changed depending on which variables the multiple linear regression models included. Reduction of B in percentage (my calculation), in the final compared to the original model, is also shown.
In the final model compared to the original model, the greatest reduction of B in percentage was seen for, in order, the variables duration (42.0%), age (24.4%), and sex (20.5%). In the original model, B for duration was .100. In the final model it was reduced to .058.
When gears was the only added variable to the original model, B for duration was lowered to .087, when only the variable showers was added, B for duration dropped to .091, and when RPE was the only addition to the initial regression model, B for duration got down to .072. Thus, while gears and showers also had an effect, RPE was the comparatively largest cause of the overall reduction of the unstandardized coefficient for duration when all the new and original variables were entered in the model simultaneously. This observation suggests that the perceived degree of exertion plays an important part in the differences in cycle commuting speed noted for varying durations. However, the fact that the variable duration still kept its significance as a predictor in the model indicates that RPE and the other added variables did not completely explain the positive relationship between cycle commuting speed and
24
Table 9. Unstandardized coefficients (B) for multiple linear regression models with the dependent variable speed and various sets of independent variables, plus reduction of B in percentage, in the final compared to the original model, and R2 for each model, for cases with a self-reported duration of ≤ 50 minutes (n = 1526)
Unstandardized coefficients B Independent variable Original
model Original model + gears Original model + showers Original model + RPE Final model (all variables) Reduction of B, in percentage, in final compared to original model Sex 2.68 2.34 2.59 2.49 2.13 20.5 Age − .086 − .077 − .082 − .074 − .065 24.4 Weight .054 .050 .053 .052 .049 9.3 BMI − .205 − .187 − .204 − .201 − .185 9.8 Duration .100 .087 .091 .072 .058 42.0
Last digit in self-reported duration − 1.45 − 1.42 − 1.46 − 1.40 − 1.39 4.1 Cycling area − 1.74 − 1.64 − 1.67 − 1.54 − 1.43 17.8 Gears 1. 73 1.55 Showers 1.51 1.31 RPE .326 .263 R2 .339 .372 .365 .359 .407
Note. How the different variables are coded or measured can be seen in Table 6.
The unstandardized coefficient for sex was also highly affected by the expansion of the regression model. This time, although showers and RPE had an effect as well, the overall reduction was to the comparatively largest extent caused by gears. This observation suggests that the amount of gears plays a part in the differences in cycle commuting speed noted for varying sexes. Moreover, because showers and RPE also were found to have an effect, it seems some of the earlier noted sex differences in relation to speed instead could be ascribed to varying access to shower facilities, as well as varying relative level of intensity.
25
6 Discussion
To my knowledge, this is the first study that analyses the relationship between cycle commuting speed and type of bicycle (in the form of varying amount of gears), access to shower facilities (showers), and rating of perceived exertion (RPE). In a multiple linear regression model also including the variables sex, age, weight, body mass index (BMI), duration, last digit in self-reported duration (1–4 or 6–9 compared to 0 or 5), and cycling area (inner urban compared to suburban and suburban – inner urban), the respondents’ expected cycling speed increased with 1.55 km/h when using a bicycle with five or more gears, with 1.31 km/h when having convenient access to shower facilities, and with 0.263 km/h for each step of higher rating of perceived exertion on the Borg RPE Scale. Each of the variables gears, showers, and RPE was found to be a statistically significant (p < .05) predictor of speed. At the same time, all the rest of the predictors, which were statistically significant in Schantz’s regression model from 2017, kept their significance also when the model was modified. Gears, showers, and RPE were also shown to play important parts in the differences in cycle commuting speed earlier noted for varying durations and sexes. All these findings combined highlights that gears, showers, and degree of perceived exertion, as well as sex, age, weight, BMI, duration, last digit in self-reported duration, and cycling area, are important when estimating speed or velocity for cycle commuters.
When interpreting the sizes of the unstandardized coefficients, it is important to bear in mind how the different variables are constructed. While the rating scale consists of a total number of 15 levels (6–20), gears and showers each consists of only two. Hence, out of these three variables, and in absolute terms, RPE could be seen as having the largest overall capacity of influence on the outcome variable, even though the expected difference tied with the change of each one unit of measure is the lowest.
The comparison between the unstandardized coefficients (B) in a regression model including all the statistically significant variables in Schantz’s model 2 from 2017 and the new model revealed each of the added variables gears, showers and RPE to be accountable for a reduction of B for duration.
26
The lowered B for duration caused by the addition of gears supports the suggestion that a higher proportion of cyclists who commute longer use more efficient (i.e. faster) bicycles. This study has also confirmed the finding by Schantz (2017), that these commuters cycle faster, as well as showed that they perceive a higher level of exertion while cycling. Put together with the findings by Hansen and Nielsen (2014), that they have higher incomes, higher levels of education, and more frequently have driving licences, these characteristics suggests that cyclists that commute longer might have some kind of special inner drive. Future studies are needed to investigate to what extent this exists.
The drop of B for duration caused by showers suggests a higher proportion of people who commute for longer durations have convenient access to shower facilities at the place of education or work. This might either be a sign of a higher frequency of shower facilities at workplaces where the workers live further away, or more likely, as some research has identified access to shower facilities to be an influential factor when it comes to choice of means of transport (Heinen et al., 2010), it is a sign of a reduced tendency to choose cycling for longer durations if the access to shower facilities is perceived as inconvenient or non-existent.
RPE was found to be accountable for the most substantial reduction of B for duration. Combined with the fact that duration kept its statistical significance even when the model was expanded, this finding provides an answer to parts of the question posed by Schantz et al. (in press; P. Schantz, personal communication, 13 December 2019) appearing earlier (in the Existing Research section): The exercise intensity in relation to the maximal oxygen uptake seems to get higher with increasing duration, although other factors (such as, presumably, higher maximal oxygen uptake) also are likely to be involved in explaining the relationship between duration and speed. Examinations of other such potential factors are proposed in studies to come.
Because the unstandardized coefficient for sex was lowered when the variable gears was added to the original regression model, the results of this study fit well with the suggestion made by Hansen and Nielsen (2014), that women generally commute on less efficient (i.e. slower) bicycles, and that this to a certain degree explains why they, on average, do not cycle as fast as men. The claim by Hansen and Nielsen (2014) that
27
women compared to men to a lesser extent use shower facilities at work, could be explained by the finding in this study that women report the access to such facilities not to be as convenient. Also, less convenient access to shower facilities was found to be another of the factors possibly explaining women’s lower average speeds. Thus, according to this study, with an equal type of bicycle (with regard to gears) and equal access to shower facilities at the place of education or work, the differences in cycle commuting speed between men and women would presumably be smaller.
In comparison with Schantz’s model, the increase of the coefficient of determination, from R2 = .34 (Schantz, 2017) to R2 = .41, together with a change of the unstandardized coefficients (B), indicate that the new, expanded model is able to describe the
relationship between the dependent and the independent variables in an even more accurate and nuanced way. However, with 41% of the variance of cycle commuting speed predictable with the current model, it still leaves 59% of the variability of this apparently complex phenomenon unaccounted for. This study was not able to control for all factors previously identified as significantly correlated with cycle commuting speed, such as the formation of the cycle paths (Boufous et al., 2018; El-Geneidy et al., 2007), the composition of types of road users, and gradient of the road (Eriksson et al., 2019; Manum et al., 2017). In addition, very likely there are some influential factors yet to be determined. To expand the model even further, and to find out more about factors governing cycle commuting speed, are suggestions for research in the future.
Regarding the use of the Borg RPE Scale, the respondents in this study showed a clear tendency of giving odd-numbered, compared to even-numbered, ratings of perceived exertion. Such rating behaviour is something I have not seen reported elsewhere, and Borg does not mention anything like it in his book (Borg, 1998). Perhaps (part of) the explanation to the tendency can be found in limited instructions about how to use the scale, combined with a preference of the respondents of giving ratings verbally anchored (which means the number is having a verbal expression, such as Somewhat hard/Hard/Very hard, attached to it)? All the odd numbers in Borg’s original RPE scale used in this study namely have such attributes, while the even numbers have not (see item 11 in PACS Q1, Appendix B).
28
6.1 Methodological considerations
With correct instructions (i.e. to follow the ones Borg [1998] refers to) and the use of the most updated Borg RPE Scale, both the odd- versus even-numbered rating bias, as well as extreme ratings such as 6 and 19, unlikely to be valid for cycle commuters on their way to education or work, might have been avoided. In future studies, this is recommended. However, as long as the deviations from the “true” ratings are evenly sized and spread throughout the entire scale, they would not, considering the way the scale has been used in this study, exert too much of influence on the interpretations of the results. In other words, if the only result of the rating bias is that a certain
percentage of the ratings have been rounded off to an adjacent, odd number, it would not be too much of a problem in this study, because it is not the exact number that is of interest, but the trends and relationship with speed.
Another consideration concerns the fact that the data collection was done a number of years ago. This could possibly have affected the results, if the behaviour of the cyclists would have changed significantly since then. The relationships between cycle
commuting speed and gears, showers, and RPE, though, are thought to be fairly consistent, even if other aspects of the cycle commuting in this particular region, such as the infrastructure or the amount of cyclists, change.
Another factor that could possibly have affected the results is the formation of the variables. Regarding gears it would have been interesting if the data had been even more differentiated. Now, due to how this particular item was designed in the PACS Q1 questionnaire, “five gears or more” formed one single category, even though it then presumably includes bicycles with very varying capacities in terms of speed. As a reference, the Stockholm County average single mode cycle commuters in the study by Schantz et el. (in press) had the following mean number of gears: 14.0 (SD = 8.7) for females, and 18.1 (SD = 7.7) for males. Therefore, next time the amount of gears is included in a study in relation to cycle commuting, an even more diversified structure of this factor could be used. Moreover, the amount of gears is likely not the only determinant of the capacity of speed for various types of bicycles. Also things such as size (including width) of the wheels, and overall weight, might be of importance. To
29
examine the relationship between cycle commuting speed and type of bicycle even deeper, even more factors in relation to bike design could be taken into account next time.
6.2 Strengths and Limitations
A key strength in this study is the sample size, which gives high statistical power. However, the respondents still constitute only a small proportion of the general population. This, combined with the fact that the sample is drawn from one particular region, implies caution regarding how to generalize.
Another strength is that this study, when examining the relationships of interest, was able to control for other factors. This measure is evidently important because the relationships between variables found in earlier studies were subject to substantial change when additional factors were controlled for. Simultaneously, this suggests the exact formation of the
relationships revealed in this study not to be completely determined. If even more factors are controlled for, the appearance of the relationships might possibly change again. Therefore, research examining probable determinants of cycle commuting speed is encouraged to
continue. Future studies are also needed to assess causation between the variables which were correlated in this study, something multiple linear regression, although suitable for the
intentions this time, is not able to do.
Finally, as this study is limited to durations ≤ 50 minutes, it lacks the possibility of extrapolating the findings for durations that goes beyond.
6.3 Conclusions
This cross-sectional study shows that cycle commuting speed is positively related to using a bicycle with five or more (as compared to four or fewer) gears, having convenient (as opposed to inconvenient or non-existent) access to shower facilities, and perceiving a higher degree of exertion while cycling. This highlights that these factors, together with sex, age, weight, BMI, duration, last digit in self-reported duration, and cycling area, are important in activities such as estimating, planning or
30
taking measures in relation to cycle commuting. Gears, showers, and RPE were also found to substantially affect the relationships between speed and duration, and speed and sex, showing that adding new factors to control for might change the earlier established relationships within this field. Finally, when the Borg RPE Scale is used, correct instructions strengthen the validity.
