The Marathon (i.e., 42.195 km or 26.219 miles) is probably the most antique endurance discipline, performed by athletes during the ancient (from the eighth century BC to the fifth century AD) and the modern (from 1894 to date) Olympic Games. Despite its history, for years this discipline was considered as an odd and potentially dangerous activity, that is, beyond the human capability to run the distance (15). However, it is now known that this preconception was unfounded, thanks in part to >40 years of scientific research (e.g., [4,6,7,18,20,26]). Several factors have been reported to influence the performance in marathon running such as the rate of aerobic metabolism, the availability of the limited stores of carbohydrate, and the velocity that can be maintained without developing excessive hyperthermia (7). The aim of marathon training is to increase the pace that can be sustained over the entire distance (13). Thus, marathon-specific research has made a significant contribution to researchers and practitioners developing new training strategies to enhance the physiological regulators of marathon performance. As a consequence, over this period, marathon performance has markedly improved; for example, the World Record was lowered from 2 hours 8 minutes 34 seconds in 1969 to 2 hours 3 minutes 59 seconds in 2010 (performed by Australian Derek Clayton and Ethiopian Haile Gebrselassie, respectively). Over the years, several studies have attempted to describe human running performance by the analysis of the decline in world-record times (1,22,29). However, surprisingly, no study has focused attention on the analysis of high-level marathon running performances across the years.
From a practical point of view, the analysis of marathon times performed in race conditions has been used to identify whether or not differences in performance are consistent between runners. However, to the best of our knowledge, only 2 studies have reported these observations. Specifically, in 1985, Sjödin and Svedenhag (25) proposed for the first time a possible classification to describe performances by male marathon runners. In this study, runners were designated into 3 different classes: (a) elite runners, with a time from 2 hours 30 minutes to 3 hours; (b) good runners, with a time inclusive among 2 hours 30 minutes and 3 hours; (c) slow runners, with a time >3 hours. In 2000, Billat et al. (4) identified 2 classes of runners: (a) top-class marathon runners, with a time from 2 hours 6 minutes 34 seconds to 2 hours 11 minutes 59 seconds; (b) high-level marathon runners, with a time from 2 hours 12 minutes to 2 hours 16 minutes. In our opinion, the aforementioned classifications may lead to a misleading interpretation of current performances. For example, using the 200 times included in the 2010 governing-body top list (i.e., the International Association of Athletics Federations [IAAF]), in the former classification (25), the whole list might be considered as elite runners (average time of 2 hours 8 minutes 54 seconds ± 1 minutes 29 seconds). Although with the latter (4), 182 and 0 runners would be top-class and high-level marathon runners, respectively. Furthermore, 18 athletes do not fit within these classifications, that is, having a time <2 hours 6 minutes 34 seconds (average time of 2 hours 5 minutes 39 seconds ± 37 seconds). Thus, these classifications seem to offer a bias description of high-level marathon runners. This may be because the authors classified the subjects involved in their studies by reflecting indirectly on the population from which subjects are sampled. On the other hand, it is useful to have a clear and detailed performance classification for both coaches and researchers to (a) analyze the marathon performance reached by high-level athletes in real settings and (b) to have an adequate time-based marathon profile to compare these results.
Therefore, the purpose of this study was to delineate a possible performance classification based on the analysis of the time progression that occurred in high-level performances in men's marathon from 1969 to 2010.
Experimental Approach to the Problem
Since the first recorded marathon time, performances have continued to improve. However, to date, no study has focused on the evolution of this performance. Furthermore, only 2 studies have proposed possible classifications to describe the performance profile of marathon runners (4,25). In our opinion, these classifications do not seem to be applicable to the entire population of high-level marathon runners. This is because there may be a biased generalization of the time-based performance profile. Therefore, there is a need for an adequate performance-based classification to describe the marathon performance profile in real settings. Consequently, the aim of this study was to define a possible performance classification of marathon events based on the analysis of the time evolution from 1969 to 2010 (i.e., 8,400 times) of high-level male marathon runners. To identify the nature of the phenomenon represented by the sequence of observations, data were collected and treated as time series (12). Once the patterns were established, the probability distribution method of the time series was used to describe the possible time range by sorting and ranking the values of the series. Thus, a possible time-based men's marathon classification was derived by using probability distribution.
The 8,400 times data were collected from 6,571 marathon specialists (mean ± SD: age 27.8 ± 3.9 years) who completed different races over a period of 42 years. Therefore, the study population involved was highly trained and ranged from national-level to international-level athletes.
Approval for the project was obtained from the local Ethics Committee. Subsequently, we analyzed the evolution of high-level men's marathon performances considering the observations collected (i.e., the time performances across the years) as a time series, that is, a sequence of data points measured at successive times spaced at uniform time intervals (12).
In this study, we analyzed the performance in men's marathon events over 42 years, from 1969 to 2010. For each date (i.e., year), observations (i.e., time performances) were gathered from official competition results available on the IAAF web site (http://iaaf.org). Furthermore, to reduce the risk of bias, the reliability of data collected was verified referring to the results available on independent-body web pages (i.e., the Association of Road Race Statisticians web site: http://arrs.net). In addition to the leading time (LT) performance per year, the whole ranking (i.e., 200 time performances, T200) was also considered so as to reduce variance between athletes across years (17).
All the data are presented as the mean ± SD. A quantitative analysis of 8,400 time performances (i.e., 200 time performances × 42 years) was performed and presented as a times-series analysis. We tested if the residuals (i.e., the difference between the observed values of the dependent variable and the values of the dependent variable that would be predicted by the model) of the linear regression were uniform using the first-order autocorrelation Durbin-Watson test (d) (8-10). Residual analysis was carried out to assess the goodness of fit or the lack of fit in the model to verify consistency of the regression estimators assuming that the error distribution is correct (14,27). When nonuniform residuals (i.e., heteroscedasticity) were observed, a polynomial regression was used, determining an nth order polynomial by using the least-squares method (i.e., the minimization of the sum of squared distances between the observed responses in the data set and the responses predicted by the linear approximation, refining the parameters by successive iterations) (12).
To determine any significant time difference from 1969 to 2010, a 1-way analysis of variance was applied to T200, after the assumption of normality was verified using the Kolmogorov-Smirnov test. When a significant F-value was found, the Tukey and Dunnet post hoc procedures were chosen in the case of equal variance being assumed or not, respectively.
The statistical analyses were performed using the software IBM® SPSS® Statistics (version 18.0.0, IBM Corporation, Somers, NY, USA). The level of significance was set at p ≤ 0.05.
For the time series analyzed (i.e., LT and T200), the first-order autocorrelation Durbin-Watson test reached values suggesting that the residuals were not independent and not normally distributed at each level of Y (d = 1.48 and 0.25, respectively). Consequently, the residual plots showed evidence of heteroscedasticity, highlighting a lack of linear regression as best descriptor. Accordingly, the least-squares method provided that the simplest polynomials which can represent the evolution of LT and T200 performance across the years were a second-order polynomial (Y = − 54,598.954 + 71.716x − 0.020x 2) and a fourth-order polynomial (Y = − 2.1E9 + 4.3E6x − 3349.476x 2 + 1.140x 3 − 0.0001x 4), respectively (Figures 1 and 2). Consequently, a significant improvement in the time required to complete the distance was observed from 1969 to 2010 for both LT and T200 (r = −0.92, p < 0.001 and r = −0.98, p < 0.001, respectively). From 1969 to 2010, the mean time differences were 2.5 ± 1.5% (i.e., 3 minutes 20 seconds ± 1 minutes 59 seconds) and 5.0 ± 2.0% (i.e., 7 minutes 1 seconds ± 2 minutes 48 seconds) for LT and T200, respectively. This corresponds to a mean improvement of 5 and 10 seconds per year for LT and T200, respectively.
Moreover, Levene's test was used to assess the equality of variance resulted in p < 0.05, rejecting the null hypothesis of equal variance. Thus, the differences obtained in sample variances are unlikely to have occurred based on random sampling; consequently, there is a difference between the variances in the population (i.e., homogeneity of variance). Accordingly, the Tukey post hoc procedures revealed that the mean T200 observed in 1969 and 1970 was significantly higher than that in the succeeding years (p < 0.009). Furthermore, the mean T200 in 2009 and 2010 was significantly lower than that in the preceding years (p < 0.002). The overall analysis suggested a disruption in marathon time improvements, indicating the presence of 3 points in the time series where the performance significantly improved with respect to that in the previous years. These points correspond to the years 1983–1984 (p < 0.001), 1997–1998 (p < 0.003), and 2003 (p < 0.001) (Figure 3).
The main finding of this study was that over the period analyzed (i.e., from 1969 to 2010), there has been a remarkable improvement in athletic ability to perform the marathon distance. Despite a nonlinear regression as best descriptor of high-level men's marathon performances across the years, the mean time tendency is oriented toward a decrement in the time necessary to complete the distance and, consequently, to an increment in running speed. In fact, a significant change throughout the period was observed for both the LT and T200 (r = −0.92, p < 0.001 and r = −0.98, p < 0.001, respectively) (Figures 1 and 2).
It is important to note that such a progression was not well distributed in the whole ranking across the various years. In fact, as highlighted in Figure 3, the presence of 3 points which characterize the overall time series has been noted. At these points, the performances significantly improved compared to those in the previous years. In particular, T200 decreased rapidly until the first point (1983–1984; p < 0.001), reaching a relative plateau until the second point (1997–1998; p < 0.003) and then started to decrease again before the third point (2003; p < 0.001). Following these observations, a new performance decrement has been noted in the last 2-year period (2009–2010, p < 0.002). The latter consideration is in line with a previous study, which speculated that the limit to performance for male marathon runners has not yet been reached (22). One possible explanation for the first drastic improvement (i.e., until the first point) is explicable by the fact that throughout this period the marathon race has evolved from an Olympic competition to a worldwide social and fitness phenomenon (for a full review, see Ref. ). Over this time, it has been documented that several events involved tens of thousands of participants (28). As a result of the increasing popularity of running marathons, new improvements in performance were achieved (i.e., until the second point). In fact, a greater number and variety of athletes from different countries had the opportunity to attempt the distance and possibly develop new training strategies (21) to improve the major physiological determinants of long-distance running performance (i.e., V̇O2max, fractional use of V̇O2max and running economy) (3). However, in our opinion, the continuous progress in the marathon performance (as highlighted by the third point) cannot be merely attributed to the advances in training methodologies. Instead, the improvement is most probably also a consequence of the drastic increase in the percentage of East-African runners (i.e., Kenyans and Ethiopians) in the marathon performance top lists (15). This phenomenon has been described by Manners (19) as ‘the greatest geographical concentration of achievement in the annals of sport’ and contributed to the marked decrease in the time required to complete a marathon race (15). For example, in the 1997 performance list (i.e., 200 times), 58 performances (i.e., 29%) were attributable to East-African runners, whereas the same analysis conducted on the 2010 performance list highlighted the presence of 167 East-African runners in the top 200 (i.e., 84%). As a result, the success of East-African runners has been extensively studied over the last decade. Studies have reported that several attributes of East-African runners such as the environment (23,24), psychological advantage (2), and favorable physiological characteristics (11,30) all contribute to achieving superior marathon performances.
With respect to the participation of East-African runners, the aim of this study was to provide a possible marathon classification based on the performance analysis from 1969 to 2010. In particular, the evidence of a third point where the performances are significantly better compared with those in the previous years (Figure 3) suggest that an updated performance rank, considering the distribution of both LT and T200 over the duration of the last 8 years of the time series (i.e., from 2003 to 2010) is required. Accordingly, to characterize this performance profile, we took into consideration the mean of LT from 2003 to 2010 to identify the upper rank, whereas the probability distribution method (i.e., quartiles) was used to assess the others ranges based on the T200 analysis. Thus, the profile of the high-level marathon performance may be classified as follows: (a) Elite Class Marathon Runners with a time ≤2 hours 5 minutes 10 seconds; (b) Top-Class Marathon Runners from 2 hours 5 minutes 11 seconds to 2 hours 8 minutes 50 seconds; (c) High Class Marathon Runners from 2 hours 8 minutes 51 seconds to 2 hours 10 minutes 10 seconds; (d) Good-Class Marathon Runners from 2 hours 10 minutes 11 seconds to 2 hours 11 minutes 10 seconds; (e) Class Marathon Runners with a time ≥2 hours 11 minutes 11 seconds. As an example, applying the aforementioned classification on the 200 times included in the 2010 IAAF top list, 5 athletes should be considered as Elite Class Marathon Runners, whereas 76, 73, and 46 should be considered as Top-, High-, and Good-Class Marathon Runners, respectively. Interestingly, all athletes assessed would fit the latter category (i.e., Good-Class Marathon Runners) confirming that the classification presented in this study may be more suitable to define the performances reached by high-level marathon runners.
Researchers and coaches engaged in performance and physiological evaluations should take into account the enhancements in the performance reported in this study. To better reflect high-level marathon performance profile, the use of time classification is recommended. Moreover, this study highlights the need to develop appropriate training strategies to elicit adaptations required for runners to be classified in the upper classes, that is, the ability to run and maintain high running velocities over the marathon distance. Thus, identifying the most appropriate training methods is fundamental in achieving improved marathon performance (e.g., high-intensity training ).
The authors would like to thank Ian Rollo (Loughborough University) for the English revision of the manuscript. Furthermore, we extend our gratitude to Andrew Drake (Leeds Metropolitan University) for his helpful suggestions concerning the manuscript. The authors dedicate this study to the General Director of the Mapei Sport Research Center, Professor Aldo Sassi, who passed away on December 12, 2010. No grant support was provided for this study. The results of this study do not constitute endorsement of the product by the authors or the National Strength and Conditioning Association.
1. Alvarez-Ramirez, J. An improved Peronnet-Thibault mathematical model of human running performance. Eur J Appl Physiol
86: 517–525, 2002.
2. Baker, J and Horton, S. East African running dominance revisited: A role for stereotype threat? Br J Sports Med
37: 553–555, 2003.
3. Basset, DR Jr and Howley, ET. Limiting factors for maximum oxygen uptake and determinants of endurance
performance. Med Sci Sports Exerc
32: 70–84, 2000.
4. Billat, VL, Demarle, A, Slawinski, J, Paiva, M, and Koralsztein, JP. Physical and training
characteristics of top-class marathon runners. Med Sci Sports Exerc
33: 2089–2097, 2001.
5. Burfoot, A. The history of the marathon. 1976—Present. Sports Med
37: 284–287, 2007.
6. Costill, DL. What Research Tells the Coach About Distance Running
. Washington, DC: American Association for Health, Physical Education and Recreation, 1968.
7. Coyle, EF. Physiological regulation of marathon performance. Sports Med
37: 306–311, 2007.
8. Durbin, J and Watson, GS. Testing for serial correlation in least squares regression. I. Biometrika
37: 409–428, 1950.
9. Durbin, J and Watson, GS. Testing for serial correlation in least squares regression. II. Biometrika
38: 159–178, 1951.
10. Durbin, J and Watson, GS. Testing for serial correlation in least squares regression. III. Biometrika
58: 1–19, 1971.
11. Esteve-Lanao, J, San Juan, AF, Earnest, CP, Foster, C, and Lucia, A. How do endurance
runners actually train? Relationship with competition performance. Med Sci Sports Exerc
37: 496–504, 2005.
12. Hamilton, JD. Time Series Analysis
. Princeton, NJ: Princeton University Press, 1994.
13. Hawley, JA and Spargo, FJ. Metabolic adaptations to marathon training
and racing. Sports Med
37: 328–331, 2007.
14. Hosmer, DV and Hjort, NL. Goodness-of-fit processes for logistic regression: simulation results. Stat Med
21: 2723–2738, 2002.
15. Larsen, HB. Kenyan dominance in distance running. Comp Biochem Physiol Mol Integr Physiol
136: 161–170, 2003.
16. Laursen, PB and Jenkins, DG. The scientific basis for high-intensity interval training
programmes and maximising performance in highly trained endurance
athletes. Sports Med
32: 53–73, 2002.
17. Lepers, R. Analysis of Hawaii ironman performances in elite triathletes from 1981 to 2007. Med Sci Sports Exerc
40: 1828–1834, 2008.
18. Loftin, M, Sothern, M, Koss, C, Tuuri, G, Vanvrancken, C, Kontos, A, and Bonis, M. Energy expenditure and influence of physiologic factors during marathon running. J Strength Cond Res
21: 1188–1191, 2007.
19. Manners, J. Kenya's running tribe. Sports Historian
17: 14–27, 1997.
20. March, DS, Vanderburgh, PM, Titlebaum, PJ, and Hoops, ML. Age, sex, and finish time as determinants of pacing in the marathon. J Strength Cond Res
25: 386–391, 2011.
21. Midgley, AW, McNaughton, LR, and Jones, AM. Training
to enhance the physiological determinants of long-distance running performance. Can valid recommendations be given to runners and coaches based on current scientific knowledge? Sports Med
37: 857–880, 2007.
22. Nevill, AM and Whyte, G. Are there limits to running world records? Med Sci Sports Exerc
37: 1785–1788, 2005.
23. Onywera, VO, Scott, RA, Boit, MK, and Pitsiladis, YP. Demographic characteristics of elite Kenyan endurance
runners. J Sports Sci
24: 415–422, 2006.
24. Scott, RA, Georgiades, E, Wilson, RH, Goodwin, WH, Wolde, B, and Pitsiladis, Y P. Demographic characteristics of elite Ethiopian endurance
runners. Med Sci Sports Exerc
35: 1727–1732, 2003.
25. Sjödin, B and Svedenhag, J. Applied physiology of marathon running. Sports Med
2: 83–99, 1985.
26. Tanaka, K and Matsuura, Y. Marathon performance, anaerobic threshold and onset of blood lactate accumulation. J Appl Physiol
57: 640–643, 1984.
27. Topp, R and Gómez, G. Residual analysis in linear regression models with an interval-censored covariate. Stat Med
23: 3377–3391, 2004.
28. Trappe, S. Marathon runners. How do they age. Sports Med
37: 302–306, 2007.
29. Whipp, BJ and Ward, SA. Will women soon outrun men? Nature
355: 25, 1992.
30. Weston, AR, Karamizrak, O, Smith, A, Noakes, TD, and Myburgh, KH. African runners exhibit greater fatigue resistance, lower lactate accumulation, and higher oxidative enzyme activity. J Appl Physiol
86: 915–923, 1999.
Keywords:Copyright © 2011 by the National Strength & Conditioning Association.
athlete; endurance; exercise performance; training