Running Performance Variability among Runners from Different Brazilian States: A Multilevel Approach

ecological model theory highlights that human development (or a given behavior) is the result of the interaction of variables derived from different levels, comprising those directly related to the subjects and those related to the environment. Given that, the purpose of this study is to establish whether runners' performance may vary among different Brazilian states, as the factors associated with this difference. The sample comprised 1151 Brazilian runners (61.8% men) that completed an online questionnaire, providing information about biological (sex, age, height, and weight), training (running pace, frequency and volume/week, and motivation), sociodemographic (place of residence and wage) aspects, and perceptions about the environmental in uences on the practice. Information about state variables was obtained from of cial institutes, and comprised the human development index (HDI), athletics events, and violence index. Multilevel analysis was conducted in

information about biological (sex, age, height, and weight), training (running pace, frequency and volume/week, and motivation), sociodemographic (place of residence and wage) aspects, and perceptions about the environmental in uences on the practice. Information about state variables was obtained from of cial institutes, and comprised the human development index (HDI), athletics events, and violence index. Multilevel analysis was conducted in HLM software. State-level characteristics explained 3% of the total variance in running performance. Of the total variance explained for the individual level, 56.4% was associated with male sex ( = 54.98;p< 0.001), age ( = 1.09;p< 0.001), body mass index ( = 6.86;p< 0.001), economic status ( = 6.23;p= 0.003), the perception of the natural environment ( = 7.58;p= 0.02), training frequency ( = 16.64;p< 0.001), and weekly volume ( = 0.30;p< 0.001). At the state level, only athletics events presented a positive and signi cant in uence on performance. There is a signi cant role of the environment on the explanation of running performance variability, and given the diversity across states, environmental variables should not be neglected, as they are relevant to the exploration of other variables possibly related to running performance. Keywords:performance; predictors; multilevel modeling 1. Introduction Sports performance is a multifactorial trait, determined by different predictors related to both the subject and the environment [1]. Previous studies have analyzed the in uence of psychological and physiological aspects [2,3], body composition [4], anthropometric traits [5], and/or environmental characteristics [6], such as training structure [7], birth- place [8], family and coaching support [9], socio-economic, and cultural aspects [10,11], which may in uence the level of sporting achievement. According to ecological model theory, human development (or a given behavior) is the result of the interaction of variables derived from different levels, organized in a hierarchical structure, comprising variables directly related to the subjects as well as those related to social, physical, and natural environments [12–14]. In this context, athletes have different life stories, train with different coaches, and live in given neighborhoods, Int. J. Environ. Res. Public Health2021,18, 3781.

levels, organized in a hierarchical structure, comprising variables directly related to the subjects as well as those related to social, physical, and natural environments [12–14]. In this context, athletes have different life stories, train with different coaches, and live in given neighborhoods, Int. J. Environ. Res. Public Health2021,18, 3781.

Int. J. Environ. Res. Public Health2021,18, 3781 2 of 12 which are located in cities/states/countries with distinct sports policies, different natural environments, and designs [15]. All of these aspects act together to produce the different athletes' pro les and performance [15,16]. These relationships can be illustrated when considering changes in runners' perfor- mance worldwide during the last decades [17]. Although there has been a signi cant increase in the number of practitioners across the years [18], improvement in runners' performance did not follow the same pattern across the continents [19]. For example, South American marathoners showed a drop of around 14% in performance, while their European peers improved by more than 40% [20]. As the largest South American country, Brazil presents nearly continental dimensions and sociocultural contrasts [21], making such differences in runners' performances notable. There is also discrepancy regarding the distribution of the best Brazilian runners across the states, with a high concentration of them located in the Southeastern region [22]. This result could be related to the existence of local sports policies that promote sport participation among its residents, thereby allowing the development of elite athletes [8,23]. Moreover, each state has different characteristics, such as population size and density; public policies, design, and infrastructure; demographic rates; human development index (HDI) [24]; and violence rates, street safety and security policies [25]; as well as speci c geographic and weather characteristics [23,26]. So, it is possible to assume that these particularities can act together to express and explain the differences observed in runners' performance [27]. Most of the studies conducted so far have primarily focused on understanding the role of individual characteristics on runners' performance [4,28–30]. Therefore, since it is already known that performance is, in part, derived from the inter-relationship between individual and environmental characteristics/contexts, this study aimed to establish whether runners' performance varies between different Brazilian states, as well as to identify potential factors associated with this performance difference, based on a multilevel modelling approach. Based on previous research that indicated differences according to the best Brazilian athletes [8,31], we hypothesized that nonprofessional runners have a signi cant

inter-relationship between individual and environmental characteristics/contexts, this study aimed to establish whether runners' performance varies between different Brazilian states, as well as to identify potential factors associated with this performance difference, based on a multilevel modelling approach. Based on previous research that indicated differences according to the best Brazilian athletes [8,31], we hypothesized that nonprofessional runners have a signi cant difference in performance. 2. Materials and Methods 2.1. Sample The sample of the present study was obtained from the InTrack project (https:// intrackproject.wixsite.com/website), a study aiming to identify factors associated with road running performance. For the present study, the sample consisted of 1151 runners (61.8% men; 38.2% women), aged between 18 and 72 years old, from 25 states and federal districts comprising the ve Brazilian regions (Southeast = 36.3%; North = 7.0%; Northeast = 35.7%; South = 12.4%; and Midwest = 8.2%). To be included in the study, runners were required to answer an online questionnaire. Participants were excluded if they had not answered all the mandatory questions from the applied questionnaire. All participants received information about the study's purposes and perspectives and gave their written consent to participate. 2.2. Instrument Information obtained from the participants was self-reported, through the question- naire section “pro le characterization and associated factors for runners' performance”, which was developed and validated previously [32]. The questionnaire provides informa- tion in six categories: (1) runner identi cation (age and sex), (2) anthropometric variables (height and weight), (3) sociodemographic pro le (neighborhood, income, educational level, and marital status), (4) perception about the environmental (natural or built) in uence on the practice, (5) training variables (volume and frequency/week, sessions/day, practice time, pace (min/km), involvement in of cial races in the least 12 months, involvement in a running club, the existence of a personal coach to guide the practice, motivation for the practice, and preferred distance), and (6) the family environment (family composition,

Int. J. Environ. Res. Public Health2021,18, 3781 3 of 12 family members engaged in running practice, involvement in sports during childhood, and family support for sporting involvement during childhood). The questionnaire was available for eligible participants using an online platform (Google forms), as used in previous studies [33–35], between November 2019 and March 2020. This online strategy was chosen to cover all Brazilian states and to maximize the variability between runners. Furthermore, it was not the purpose of the study to obtain a representative sample in each of the Brazilian states, nor nationally. 2.2.1. Individual-level variables Biological variables Sex, age, body height, and body weight were self-reported. Body mass index (BMI) was computed by the standard formula: weight (kg)/height (m 2 ). Training variables Running pace: Running pace was used as the primary performance indicator (included in the model as the outcome variable). Runners were asked to state their run pace in preferred distance. Frequency of training: Runners were asked to state the number of training sessions they complete per week (1–7 train/week). The variable was dichotomized as either at least 3 sessions/week or more than 3 sessions/week. Volume/week: Runners were asked to provide information about the average total distance (in kilometers) they usually cover during their weekly training sessions. 2.2.2. Sociodemographic Socioeconomic status (SES): Runners were asked to provide an estimate of their monthly income, in a Likert scale format, based on Brazilian minimum wage in 2019 [36]. Answers were restructured in the following categories: low ( BRL 998.00 or about < USD 241.06), medium (> BRL 998.00– BRL 2994.00 or about USD 241.06– 723.18), medium–high (> BRL 2994.00– BRL 4990.00 or about > USD 723.18– USD 1205), and high (> BRL 4990.00 or about > USD 1205), which were used in the analysis. Place of residence. Runners were asked about the city they live in (state capital or not). Perception of the environmental in uences for the practice Weather: Runners were asked about their perception of the in uence of the natu- ral environment (namely weather conditions) during running practice. Based on their answers, the variable was dichotomized

were used in the analysis. Place of residence. Runners were asked about the city they live in (state capital or not). Perception of the environmental in uences for the practice Weather: Runners were asked about their perception of the in uence of the natu- ral environment (namely weather conditions) during running practice. Based on their answers, the variable was dichotomized to yes (it in uences) or no (it does not in uence). Physical structures: The perception about the presence of physical structures and the environment (the existence of parks/places for the practice, street safety and design), that can promote ongoing running practice, was obtained and dichotomized to yes (it in uences) or no (it does not in uence). 2.2.3. State-level variables Information was obtained from of cial institutes, such as the Atlas of Human Devel- opment in Brazil [37], the Atlas of Violence [25], and the Brazilian Institute of Geography and Statistics [21] for each state. Human development index (HDI): Based on the HDI, states were categorized as medium ( 0.699), high ( 0.700 and 0.799), or very high ( 0.800) HDI. None of the states had an HDI classi ed as low (<0.600). Athletics events: Information regarding the existence of athletics events in the various Brazilian states was obtained from State Basic Information Research [38]. The variable was categorized as either yes (there is) or no (there is not).

Int. J. Environ. Res. Public Health2021,18, 3781 4 of 12 Violence index: Femicide was used as the violence index indicator, obtained from the Atlas of Violence [25]. It expresses the total number of women homicides by year in each state. 2.3. Statistical analysis Descriptive statistics are presented as a mean standard deviation (SD) and fre- quencies, and were computed using IBM SPSS Statistics (IBM Corp. Released 2016. IBM SPSS Statistics for Windows, Version 26.0. Armonk, NY: IBM Corp). The running pace was considered the outcome variable, and the Hierarchical Linear Model (performed in software HLM 6.0) was computed to estimate the association of individual- and state-level variables and performance variance. A series of hierarchically nested models were tted, and the model accuracy was analyzed based on the deviance statistics value, which is expected to signi cantly decrease as the model complexity increases, and this decrease is tested by a chi-square test [39]. After that, the relevance of the predictors included was determined through a pseudo-R 2 statistical test, which was interpreted as the proportion of the variance reduction for the parameter estimate, which is a result from the comparison of a given model to its previous one [40]. Further, models were built in a stepwise fashion, as generally suggested [39,41]: the rst step of the analysis comprised the running of the null model, which allowed computing of the intracluster correlation coef cient to estimate the variance that accounted for states effects on performance. Secondly, the individual-level model (model 1), with the inclusion of the subject predictors (sex; age; BMI; training variables, i.e., frequency and volume/week; sociodemographic factors (SES; place of residence) and individual perception about the in uence of the natural and built environment on their practice) was run. Thirdly, the state-level model (model 2) was computed with the insertion of the state-level variables (HDI, violence index, and athletics events). For all the analyses, the signi cance level was set atp< 0.05. 3. Results The sample consisted of runners from both sexes, aged between 18 and 72 years. A total of 61.8% of runners were classi ed as normal weight

the state-level model (model 2) was computed with the insertion of the state-level variables (HDI, violence index, and athletics events). For all the analyses, the signi cance level was set atp< 0.05. 3. Results The sample consisted of runners from both sexes, aged between 18 and 72 years. A total of 61.8% of runners were classi ed as normal weight according to their BMI. The run- ners reported an average pace slightly lower than 5:30 min/km, and had a heterogeneous weekly distance covered per week (35.49 29.54 km). The majority of the participants re- ported practicing for more than one year, and training up to three sessions/week. Runners most frequently lived in state capitals, in states with medium HDI, and with a wide range of feminicide cases per year. Almost 95% of runners received a monthly income above and up to ve times the Brazilian minimum wage. Data suggested that it is common the promotion of athletics events in states. Moreover, both the natural and built environment seemed to in uence running practice (Table). The nal estimated variance at the state level, presented in the null model, was found to be signi cant, revealing statistical differences across the states. The intracluster correlation coef cient showed that ~3% of the total variance in runners' performance was explained by the differences between states, meaning that 97% of this variance is explained by runners' individual characteristics (Table). Figure the medians, across Brazilian states. It is possible to observe differences between states, ranging from 252 to 360 s/km, as well as relevant within-state differences.

Int. J. Environ. Res. Public Health2021,18, 3781 5 of 12 Table 1. Descriptive statistics (means and standard deviation; or frequency) of the individual- and state-level variables. Variables Mean (Standard Deviation) or Frequency (%) Sex Male 711 (61.8%) Female 440 (38.2%) Age (years) 37.9 (9.4) BMI (kg m 2 ) 24.3 (3.1) Practice time 1 year 173 (15.0%) >1 year 976 (84.8%) Running pace (s) 324.2 (57.7) Volume training/week (km) 35.5 (29.5) Frequency training/week 3 sessions/week 678 (58.9%) >3 sessions/week 473 (41.1%) Live in capital No 512 (44.5%) Yes 639 (55.5%) Socioeconomic status (SES) Low 65 (5.6%) Medium 542 (47.1%) Medium-high 526 (45.7%) High 4 (0.3%) Natural environment in uences No 371 (32.2%) Yes 779 (67.7%) Physical environment in uences No 299 (26.0%) Yes 852 (74.0%) Athletics events No 5 (19.2%) Yes 21 (80.8%) Human development index Medium 13 (50.0%) High 12 (46.2%) Very High 1 (3.8%) Female homicides 187.4 (150.1)Int. J. Environ. Res. Public Health 2021, 18, x 5 of 13 Figure 1. Running pace variance between Brazilian states, sorted by the median. Box-plot elements are as follow: cross, median; superior and inferior box limits, mean 75th and 25th percentiles, re- spectively; top and bottom bars, maximum and minimum normal values, respectively; circles indi- cate outliers. Table 1. Descriptive statistics (means and standard deviation; or frequency) of the individual- and state-level variables. Variables Mean (Standard Deviation) or Frequency (%) Sex Male 711 (61.8%) Female 440 (38.2%) Age (years) 37.9 (9.4) BMI (kg·m −2 ) 24.3 (3.1) Practice time ≤1 year 173 (15.0%) >1 year 976 (84.8%) Running pace (s) 324.2 (57.7) Volume training/week (km) 35.5 (29.5) Frequency training/week ≤3 sessions/week 678 (58.9%) >3 sessions/week 473 (41.1%) Live in capital No 512 (44.5%) Yes 639 (55.5%) Socioeconomic status (SES) Low 65 (5.6%) Medium 542 (47.1%) Medium-high 526 (45.7%) High 4 (0.3%) Natural environment influences No 371 (32.2%) Yes 779 (67.7%) Physical environment influences Figure 1. Running pace variance between Brazilian states, sorted by the median. Box-plot elements are as follow: cross, median; superior and inferior box limits, mean 75th and 25th percentiles, respectively; top and bottom bars, maximum and minimum normal

(5.6%) Medium 542 (47.1%) Medium-high 526 (45.7%) High 4 (0.3%) Natural environment influences No 371 (32.2%) Yes 779 (67.7%) Physical environment influences Figure 1. Running pace variance between Brazilian states, sorted by the median. Box-plot elements are as follow: cross, median; superior and inferior box limits, mean 75th and 25th percentiles, respectively; top and bottom bars, maximum and minimum normal values, respectively; circles indicate outliers.

Int. J. Environ. Res. Public Health2021,18, 3781 6 of 12 Table 2.Summary of results of the hierarchical linear model for the variance in running performance. Parameters Null Model Model 1 Model 2 Estimates Standard Error p-Value Estimates Standard Error p-Value Estimates Standard Error p-Value Intercept 323.65 2.92 <0.001 346.49 5.81 <0.001 356.81 5.36 <0.001 Sex 54.98 3.08 <0.001 55.25 3.13 <0.001 Age 1.09 0.12 <0.001 1.12 0.12 <0.001 BMI 6.86 0.52 <0.001 6.88 0.52 <0.001 Place of residence 0.02 3.17 0.995 0.15 3.30 0.963 SES 6.23 2.06 0.003 6.42 2.01 0.002 Natural environment 7.58 3.25 0.02 7.50 3.26 0.022 Physical structure 3.89 2.71 0.152 3.96 2.68 0.140 Frequency/week 16.64 2.65 <0.001 16.45 2.55 <0.001 Volume/week 0.30 0.08 <0.001 0.30 0.08 <0.001 Athletic events 9.36 2.34 0.001 Woman homicides 0.01 0.01 0.139 HDI 5.21 2.93 0.089 Variance components: random effects Between-states 104.13 45.34 9.79 Within-sates 3261.73 1479.73 1484.38 Model summary Deviance statistic 12,595.210 10,093.54 100,85.06 Number of estimated parameters 3 12 15 BMI, Body Mass Index; SES, Socioeconomic status; HDI, Human Development Index.

This study analyzes performance variability among Brazilian runners based on various factors.