Fast men slow more than fast women in a 10 kilometer road race

Statistical analysis

We considered two measures of pacing. For our first pacing measure, we calculated the percentage change in the pace maintained over the first 3 miles relative to the final 3.2 miles. That is, % change = (minutes per mile in final 3.2 miles − minutes per mile in first 3 miles)/minutes per mile in first 3 miles. We used this measure because it is similar to comparing the pace of the first half of the race to the second half of the race, which has been done in marathon studies (Deaner et al., 2015a; Krawczyk & Wilamowski, 2015). (No 5 km split was available for this race, preventing a more direct comparison with previous studies.) Our second pacing measure was the percentage change in the pace of the first mile relative to the final 5.2 miles. That is, % change = (minutes per mile in final 5.2 miles − minutes per mile in first mile)/minutes per mile in first mile. We selected this measure because it should capture runners beginning the race at a speed which is too fast to be sustainable.

We considered each of these two pacing measures as a continuous variable and as two categorical variables. For the first categorization, percentage changes less than 5% were considered “maintaining the pace.” For example, a runner who completed the first 3 miles of the race in 6 min and 40 s per mile and the final 3.2 miles in 7 min per mile or better maintained the pace; those who ran the final 3.2 miles in more than 7 min per mile failed to maintain the pace. For the second categorization, percentage changes equal to or greater than 10% were considered as “marked slowing.” For example, a runner who completed the first 3 miles of the race in 6 min and 40 s per mile and the final 3.2 miles in 7 min and 20 s per mile or worse showed marked slowing. Total finishing time and all mile split times were determined from electronic chip times (i.e., based on when each individual crossed the starting line).

To explore the impact of finishing time on our categorical pacing measures, we also categorized male and female performances into sex-specific sextiles (i.e., dividing all performances of one sex into six equivalent sized groups). The fastest male group consisted of men finishing in 48:40 or faster, and the other sextile boundaries for men were 53:54, 58:30, 63:40, and 71:05. The corresponding boundaries for women were 55:27, 60:28, 64:58, 69:53, and 75:41.

When considering pacing as a continuous response (or dependent) variable, we estimated two multiple regression models for each of our two pacing measures. Model 1 used sex, age, and finishing time as explanatory (or independent) variables. Model 2 included sex, age, and finishing time, along with the pairwise interactions between these variables. For these models and some figures, we followed a previous study (Deaner et al., 2015a) in dividing women’ times by 1.12 to account for men’s greater maximal oxygen uptake (Joyner, 1993; Sparling, O’Donnell & Snow, 1998; Cheuvront et al., 2005).

Descriptive summaries are presented as mean ± standard deviation (SD). All analyses were conducted using the R programming language, version 3.1.1 (R Core Team, 0000). Data are reported as mean ± SD in the text. Two-sided P-values less than 0.05 were taken as statistically significant.

Pacing by sex and finishing time

Table 1 reports the results for the two pacing measures (Measure 1: first 3 miles relative to the final 3.2 miles; Measure 2: first mile relative to the final 5.2 miles) for each finishing time sextile, separately for men and women. For both measures and both sexes, change in pace varied significantly across sextiles (P < 0.0001 for both sexes and both measures), with slower runners generally experiencing greater slowing. Among runners in the fastest two sextiles, men slowed significantly more than women for both pacing measures. Among the other sextiles, women sometimes slowed significantly more than men (fourth sextile, measures 1 and 2; fifth sextile, measure 2), men sometimes slowed significantly more than women (third sextile, measure 1; sixth sextile, measures 1 and 2), and in some cases, there was no significant sex difference (fifth sextile, measure 1; third sextile, measure 2).

We next determined, for each pacing measure, the percentage of men and women in each sextile that maintained the pace (<5% slowing). Table 2 shows that, for both pacing measures, maintaining the pace varied across sextiles, as faster runners were more likely than slower runners to maintain the pace (P < 0.0001 for both sexes and both measures).

We calculated the common (pooled) OR for maintaining the pace for women compared with men across the sextiles (Table 2). Overall, the odds that women maintained the pace were 1.16 times (95% CI [1.13–1.18]; P < 0.0001) greater compared to men for the first pacing measure. For the second pacing measure, the odds that women maintained the pace were 0.97 times as great as the odds that men maintained the pace (95% CI [0.95–0.99]; P < 0.0001). For both measures, there were significant differences among the sextiles (P < 0.0001 for both measures, indicated by a Breslow–Day test for homogeneity). The largest sex differences occurred in the fastest sextile (first measure: OR, 1.96; 95% CI [1.82–2.10]; second measure: OR, 1.36; 95% CI [1.30–1.42]).

We conducted similar analyses of marked slowing (≥10% slowing). Table 3 shows that, for both measures, marked slowing varied across sextiles, as faster runners were less likely to exhibit marked slowing (P < 0.0001 for both sexes). Overall, the odds of exhibiting marked slowing were 0.79 (95% CI [0.76–0.81]; P < 0.0001) times as great for women compared with men for the first pacing measure, and 0.98 times (95% CI [0.97–1.00]; P = 0.10) as great for women compared with men for the second pacing measure (Table 3). There were significant differences among the sextiles (P < 0.0001 for both measures). The greatest sex differences occurred in the fastest sextile (first measure: OR, 0.46; 95% CI [0.38–0.55]; second measure: OR, 0.65; 95% CI [0.61–0.68]).

To further explore whether the relationship between finishing time and pacing differed for men and women, we plotted these relationships separately for each sex. In these plots, we adjusted (i.e., decreased) female finishing times by 12% (see ‘Methods’). Figure 1 illustrates the results, (A) Pacing Measure 1, and (B) Pacing Measure 2. This figure indicates that, for both measures, slower finishers exhibited greater percentage slowing, although this trend was most prominent for those who finished in approximately 60 min or more. In addition, this figure shows that, for both measures, men tended to slow more than women among the fastest runners (i.e., roughly those who finished in under 53 min, corresponding to a time of 59:36 for women). Among slower runners, there was no consistent sex difference in pacing for the first pacing measure (Fig. 1A). For the second pacing measure, among slower runners, women showed a consistent tendency to slow more than men (Fig. 1B). We also plotted these relationships without making a 12% adjustment to women’ finishing times (Fig. 2). With these absolute finishing times, men slowed more than women among faster and slower runners, and this was true for both pacing measures.

Regression modeling to control for runner’s age

The preceding results investigated the relationships among pacing, sex and finishing time through tabular and visual displays. In order to simultaneously address the potential impact of runner’s age, we used multiple regression modeling. We fit two models for each of our two continuous pacing measures. Model 1 used sex, age, and finishing time as explanatory variables. Model 2 additionally included the pairwise interactions between sex, age, and finishing time. In these models, finishing time was measured in minutes (adjusted by 12% for women) and age was measured in years.

For Pacing Measure 1, older, faster, and female runners exhibited more even pacing (Model 1). With regards to sex differences, Model 2 indicated that the effect of sex was attenuated for slower finishers. That is, men generally showed greater slowing than women, but this difference weakened as finishing time increased.

For Pacing Measure 2, older and faster runners showed more even pacing, while males and females were not significantly different (Model 1). The lack of significance of the sex variable in Model 1 reflects a significant interaction between sex and finishing time (Model 2). That is, among fast runners, men showed greater slowing but, among slower finishers, women showed greater slowing.

These regression results corroborated the patterns illustrated in Fig. 1. Complete information on the value of the coefficients and standard errors from these models are shown in Table 4.

Susceptibility to glycogen depletion hypothesis

The documentation of a sex difference in pacing in a 10 km road race challenges the hypothesis that the sex difference in pacing is due to men’s greater susceptibility to glycogen depletion (March et al., 2011; Trubee et al., 2014). According to that hypothesis, no sex difference in pacing should have occurred for a 10 km race because glycogen depletion is highly unlikely to be relevant for events of this duration (Coyle, 2007; Rapoport, 2010). Nevertheless, it is premature to completely reject the susceptibility to glycogen depletion hypothesis because the present study of a 10 km race cannot discount that men’s apparently greater susceptibility to glycogen depletion contributes to the sex difference in pacing in marathons. The fact that the sex difference in pacing in this 10 km race was limited to faster runners and was smaller in magnitude than the sex difference in marathon pacing is consistent with this possibility.

Decision making hypothesis

The present study’s results support, albeit indirectly, the hypothesis that the sex difference in pacing in distance running reflects, at least in part, some kind of sex difference in decision making (Allen & Dechow, 2013; Deaner et al., 2015a; Krawczyk & Wilamowski, 2015; Hubble & Jinger, 2015). For example, men may be more likely than women to begin a race with an ambitious or risky pace relative to their ability, and this would increase their likelihood of slowing later.

The decision making hypothesis is supported by several lines of evidence besides the present study’s results. First, men’s greater slowing in the Warsaw (Krawczyk & Wilamowski, 2015) and Houston marathons (Hubble & Jinger, 2015) was substantially associated with their more discrepant forecasting, indicating men were typically more ambitious or over-confident than were women. Second, a study of the Chicago marathon showed that performances among non-elite runners clustered at round numbers (e.g., just under 4 h) because some runners accelerated in the final 2.2 km of the race; although women were more likely than men to have accelerated in the final 2.2 km, men were more likely to have done so in order to achieve round numbers, indicating that the men were more likely to modify their pace to pursue external goals (Allen & Dechow, 2013). Third, in non-sport domains, men generally take greater risks, partly due to their lesser perceptions of risk (Harris, Jenkins & Glaser, 2006), and laboratory studies of running and cycling indicate that individuals with lesser perceptions of risk are more likely to adopt ambitious initial paces (Micklewright et al., 2015). Finally, risk taking and competitiveness are believed to be closely related (Croson & Gneezy, 2009), and there is mounting evidence that male distance runners are typically more competitive than their female counterparts (Deaner, 2013; Deaner, Addona & Mead, 2014; Deaner et al., 2015b).

Other physiological factor(s) besides glycogen depletion may contribute to the sex difference in pacing. The only suggestion of which we are aware is that the sex difference in pacing might be due to men’s (supposedly) greater susceptibility to hyperthermia (Trubee et al., 2014), a frequent contributor to slowing in distance running (Maughan, Watson & Shirreffs, 2007; Gonzalez-Alonso, Crandall & Johnson, 2008; El Helou et al., 2012). This hypothesis was suggested by the fact that the sex difference in marathon pacing increases with warmer ambient temperatures (Trubee et al., 2014; Krawczyk & Wilamowski, 2015). The present study’s results from the Bolder Boulder 10 km race do not address the susceptibility to hyperthermia hypothesis. (We did not address the effect of temperature because there was little variation in the race years we studied; the maximum daily temperature in Boulder on race day was 23.06 ± 3.31 °C.) We note, however, that this hypothesis is weakened by the fact that, compared to women, men are generally believed to have advantages in thermoregulation (Gagnon & Kenny, 2012). Moreover, the sex difference in pacing has been documented in marathons (March et al., 2011; Trubee et al., 2014; Krawczyk & Wilamowski, 2015) that occurred at temperatures sufficiently cool (e.g., <10 °C) that hyperthermia would have been irrelevant (El Helou et al., 2012).

Another crucial point is that although there is, at present, no compelling evidence that a physiological difference between men and women can explain the sex difference in pacing, the decision making hypothesis is not at odds with physiology contributing to individual and group variation in pacing. This is because the decision to adopt an ambitious or risky pace will undoubtedly impose physiological challenges (Coyle, 2007; Abbiss & Laursen, 2008; Tucker & Noakes, 2009).

Why men’s greater slowing was limited to faster runners

An interesting question is why the greater slowing by men than women in this 10 km road race was limited to relatively fast runners, roughly men who finished in less than 53 min and women who finished in less than 60 min. By contrast, the sex difference in marathon pacing occurs for runners of widely ranging abilities (March et al., 2011; Trubee et al., 2014; Deaner et al., 2015a; Krawczyk & Wilamowski, 2015). Furthermore, research on sex differences in risk taking in non-sports domains (e.g., driving) generally indicate that sex differences occur across broad populations (Harris, Jenkins & Glaser, 2006; Croson & Gneezy, 2009) or, if they vary across populations, the sex differences tend to be smaller or non-existent among more selective sub-populations (Croson & Gneezy, 2009 but see Deaner et al., 2015b). The sex difference in financial risk taking, for instance, apparently weakens or disappears among professional investors (Croson & Gneezy, 2009).

Although we cannot definitively explain why the sex difference in pacing was limited to relatively fast runners, we can offer a suggestion, based on two points. First, although pacing in distance running can involve risk taking, it often may not. For instance, an evenly paced performance may not indicate a runner whose gamble with an ambitious risky pace paid off; it may simply indicate a runner who adopted a pace that was easy for them. Conversely, a performance with dramatic slowing may not indicate a runner who began the race with an ambitious pace but, despite great effort, failed to hold it; it may merely indicate a runner who did not prioritize achieving an outstanding performance and was content to slow when their initial pace became uncomfortable. Second, although any particular running performance may or may not involve substantial risk taking, it can be expected that faster runners generally will be more likely than slower runners to select an ambitious pace and to exert themselves in maintaining it. This is supported by surveys showing that faster runners are more likely than are slower runners to endorse competitiveness and goal achievement items (e.g., “to push myself beyond my current limits”) (Masters, Ogles & Jolton, 1993; Deaner et al., 2015b). If these points are correct, then variation in pacing among fast runners may provide important insights about risk taking, whereas variation in pacing among slower runners may be less informative.

Our suggestion for why the sex difference in pacing was limited to faster runners in this 10 km race does raise the question of why the sex difference in pacing in marathons held even among slower runners (March et al., 2011; Trubee et al., 2014; Deaner et al., 2015a; Krawczyk & Wilamowski, 2015). Although we cannot provide a definitive answer to this question, we can suggest that 10 km races and marathons may differ in ways that make marathon races more likely to reveal individual and group variation in decision making. One difference is that somewhat different kinds of individuals may participate; marathoners tend to be more motivated by competition (Ogles, Masters & Richardson, 1995). A second difference involves the demands of maintaining an even pace, a reliable correlate of performance in distance running (Abbiss & Laursen, 2008; Tucker & Noakes, 2009; March et al., 2011; Trubee et al., 2014; Santos-Lozano et al., 2014; Deaner et al., 2015a; Krawczyk & Wilamowski, 2015; Hubble & Jinger, 2015; Hanley, 2016). Because fatigue in a marathon generally reflects slowly cumulating processes, such as glycogen depletion (Coyle, 2007; Rapoport, 2010), hyperthermia (Maughan, Watson & Shirreffs, 2007; Gonzalez-Alonso, Crandall & Johnson, 2008; El Helou et al., 2012), and muscle damage (Del Coso et al., 2013), coaching handbooks advise that achieving an evenly paced marathon requires running the first half of the race at a pace that seems sufficiently comfortable that the runner is frequently tempted to accelerate in order to “bank some minutes” and get ahead of their time goal; resisting this temptation may be crucial for avoiding dramatic slowing late in the race (Humphrey, Hanson & Hanson, 2012; Magness, 2014). By contrast, in shorter races, such conservatism seems less important, apparently because immediate feelings of discomfort are generally sufficient to inform the runner that they should reduce their speed to avoid severe distress. Underscoring the unusual pacing demands in the marathon is that the mean percentage slowing in marathons (Deaner et al., 2015a) is far greater than documented in this 10 km race.

Future work on decision making and pacing

Although the present study’s results, in conjunction with previous research, indicate that decision making contributes to the sex difference in pacing, additional research must be conducted to make strong tests of this hypothesis. Ideally, this work would investigate when and how runners establish their pre-race goals, including their target pace(s); runners’ estimates of the benefits, costs, and likelihoods of various race outcomes (e.g., dramatic slowing, setting a personal best, winning age group); runners’ psychological and physiological experiences during the race as they adjust their pace and modify their goals; and how these factors vary according to sex, age, experience, risk taking, training, ability, coaching, course difficulty, and race distance. Besides allowing a strong test of whether decision making contributes to the sex difference in pacing, such research may reveal unexpected and interesting interactions among physiological and psychological factors.

If it turns out that the sex difference in pacing partly reflects a sex difference in some aspect of decision making (e.g., over-confidence, risk perception, willingness to tolerate discomfort), then coaches and athletes may benefit by considering individual and group variation in these characteristics when planning training and racing. For example, training sessions aiming to improve pacing for males (or more confident females) may need to emphasize being conservative whereas corresponding sessions for females (or less confident males) may require encouraging more ambitious pacing.

Limitations

The current study has several limitations. First, we evaluated pacing by comparing the pace of the first 3 miles to the final 3.2 miles and by comparing the pace of the first mile relative to the final 5.2 miles. Although these two pacing measures produced roughly similar outcomes with respect to sex differences among faster runners, other pacing measures could be explored and might yield different results. Second, although we succeeded in identifying sex, finishing time, and age as contributors to pacing variation, other factors must also be important. For example, in the study of 14 US marathons, greater race experience was associated with lesser slowing, although the effect was modest and did not eliminate the sex difference in pacing (Deaner et al., 2015a). As noted above, other relevant factors might include risk taking, training, and coaching. Third, although our sample size was large in terms of runners, the sample was drawn entirely from six years of the same race, the Bolder Boulder 10 km road race. This may limit the generalizability of our conclusions. Finally, many runners participated in more than one year of the race, and we did not cluster performances by runner. This could have led to p-values being smaller than they would have been if clustering was modeled explicitly.