Physical education environment and student physical activity levels in low-income communities

This study used data collected for a single-county evaluation project in Illinois, U.S.A. that evaluated student PA levels during PE classes in public schools. The geographic area served by this evaluation is considered to be a suburban area with approximately 2.5 million people (57% non-Hispanic white, 19% Hispanic, 16% black, and 16% Asian). Most public schools in this suburban area were organized into elementary, middle, and high schools. The state of Illinois had a daily PE requirement, although waivers and exemptions were allowed. The state of Illinois had no mandatory policy, but provided recommendations, regarding a PE specialist and curriculum. The evaluation focused on nine school districts in the county with ≥40% of students eligible for free- or reduced-price lunch. All nine districts were invited to and agreed to participate in the evaluation. The evaluation involved six waves of SOFIT assessments over two school years (fall 2015, early winter 2016, late winter 2016, spring 2016, fall 2016, and spring 2017). Among the total of 43 schools within the nine districts, three schools (7%) were disengaged from the evaluation during the scheduling period for the first or second assessment. Of the remaining 40 schools, 26 were elementary schools, six were middle schools, and eight were high schools. After the first assessment in fall 2015, schools were offered various supports, such as PE teacher training, school administrator training for PA and PE policies, and technical assistance for policy improvement. Trainings were designed to provide the information needed for school staff to make improvements to their PE policies, as well as capacity building skills and strategies for addressing barriers in the improvement process. Although PE teacher’s participation in the PE training was voluntary, the majority of PE teachers participated in the training. As we found that the wave 1 data had a large proportion of missing data or outliers (> 10%) and student PA levels estimated in wave 1 were substantially different from those in waves 2 to 6, we excluded the wave 1 data and used the waves 2 to 6 data for this report.

At each wave, two typical school days per school were selected to conduct the SOFIT observations. All PE classes scheduled on those 2 days were observed. The number of PE classes per day varied, ranging from two to eight. A total of 2106 PE classes across the 40 schools were observed. For the current analysis, we excluded six PE classes that were solely for students with special needs, 24 PE classes that were for preschoolers, and 14 PE classes that were observed for less than 16 min, resulting in 2063 PE classes.

The System for Observing Fitness Instruction Time (SOFIT) was used to assess student PA, lesson context, and teacher behavior. The SOFIT is a direct observation method that utilizes a momentary time sampling and interval recording system to assess PA levels, lesson context, and teacher behavior over a designated class period [15]. SOFIT evaluates student activity level in five categories: lying down, sitting, standing, walking and vigorous activity. Lesson context is categorized into general content (transition, management, and break), knowledge content (physical fitness, general knowledge, rules, strategy, social behavior, and technique), and motor content (fitness, skill practices, game play, and other). Teacher behavior is classified into three categories: when in-class PA or fitness is promoted (in-class PA promotion), when out-of-class PA or fitness is promoted (out-of-class PA promotion), and when neither in- nor out-of-class PA or fitness is promoted (no PA promotion). The definitions of all these categories can be found in the SOFIT manual [16]. The guidelines for SOFIT study reporting suggested by McKenzie and Smith [11] informed the data reporting of this report. The observer training process was conducted in accordance with the SOFIT manual [16]. Briefly, a project manager with extensive experience on SOFIT observations led the observer training. Observers completed classroom training, video analysis, and field practice. Classroom training included a general overview of the method, a detailed review of the protocol and question and answer sessions. The observers watched the SOFIT training and coding practice videos suggested in the manual. Finally, the observers observed a youth soccer practice. After the observation, they debriefed the experience. Discussion included a review of any data points for which codes between observers were discordant. The observers discussed and came to consensus on the appropriate code for that situation and notes were added to the internal protocol manual for consistent application of the code during data collection.

Five waves of SOFIT assessments were conducted in early winter 2016 (January and February), late winter 2016 (March to early April), spring 2016 (late April and May), fall 2016 (September and October), and spring 2017 (April and May). Research staff contacted a school representative and scheduled assessment visits (2 days per school at each wave). Only typical school days (e.g., no field day or no shorter day) were selected for the SOFIT assessment. Those 2 days were mostly two consecutive days. However, when scheduling for two consecutive days was not allowed, 2 days that were not on the same day were selected within 2 weeks (e.g., Monday and following week Tuesday). The research staff were given the PE class schedules for the scheduled assessment days. One observer conducted the SOFIT observation per PE class. At the onset of the PE class, four target students and one potential target student as a backup were selected to assess student PA. The observer observed student one for 4 min, then rotated the observation to the remaining three students for 4 min each until the lesson ended. The observer recorded student activity, lesson context, and teacher behavior every 20 s for the entire PE class, using a paper-form of the SOFIT coding sheet. The paper-form data were entered into an electronic database system that automatically summarized activity, lesson context, and teacher behavior per student and per class. For data quality control, data entry for 10% of the observed PE class sample was cross-checked by independent research staff.

In addition to the direct observation of student PA, the observer recorded descriptive characteristics of the PE class, such as grade level, lesson location (indoor, outdoor, or both), scheduled length of PE class, the number of PE teachers, and the number of students. All observed PE classes were gender-mixed classes, with approximately half males and half females. The exact class gender composition was not recorded. Data on school characteristics such as racial/ethnic composition of the student body and percentage of low-income students (determined by students who are eligible for free- or reduced-price lunch) were obtained from the www.​illinoisreportca​rd.​com website provided by the Illinois State Board of Education.

To check for potential data entry errors, extensive descriptive statistics, including range, median, mean, and outlier tests, per class, per day, per school, and per wave were conducted. An activity level was evaluated as a percentage of PE class time (frequency) spent in the activity level. For example: If walking activity was recorded for 30 intervals during a 90-interval (that is, 30-min) observation, walking activity level was 33% (calculated as 30 divided by 90). Because the lying down activity was rare, we combined lying down and sitting. Walking and vigorous activities were also combined as MVPA. MVPA was then dichotomized as ≥50 and < 50%. Teacher behavior categories and lesson context categories were also evaluated as percentages in a similar manner. Lesson time spent in each of in-class PA promotion, motor content, game play content, and fitness content was dichotomized as higher and lower. The cut-points for dichotomization were determined based on the median value in PE classes with MVPA≥50%. Specifically, for in-class PA promotion, “higher” was defined as ≥20% for elementary school, ≥15% for middle school, and ≥ 10% for high school. For motor content, “higher” was defined as ≥60% for elementary school, ≥65% for middle school, and ≥ 70% for high school. For game play content, “higher” was defined as ≥20% for elementary and middle schools and ≥ 30% for high school. For fitness content, “higher” was defined as ≥30% for all school levels.

To calculate the number of students per PE teacher (student-teacher ratio), the total number of students was divided by the number of teachers in class. For the lesson location variable, “outdoor” and “both” categories were combined as “outdoor,” because only < 5% of the PE class sample occurred in “both” and MVPA levels for “outdoor” and “both” were similar. To assign a single numeric grade value for grade-mixed classes, a median value was used. For example, the grade value for a first and second grade-mixed PE class was 1.5. The grade value for Kindergarteners was assigned as 0. The number of male students observed (1, 2, 3, or 4) was calculated. School race/ethnicity was assigned based on the majority race/ethnicity of the student body (black, Hispanic, white, or other).

All statistical analyses were conducted separately by school level: elementary, middle, and high school. Means and 95% confidence intervals (CIs) of the main study variables were calculated. To examine the relationship of PE class characteristics and school characteristics with MVPA (≥50% or < 50%), chi-square tests and t-tests were conducted. To account for within-school random effects, mixed models were used to predict MVPA ≥50%. The first set of mixed logistic regression models (Model 1) included lesson location (outdoor vs. indoor), in-class PA promotion (high vs. low), and motor content (high vs. low) as the predictors of interest. To examine more detailed motor content of the PE lessons, the second set of mixed models (Model 2) included lesson location, in-class PA promotion, fitness content (high vs. low), and game play (high vs. low) as the predictors of interest. All mixed models were adjusted for majority race/ethnicity of the student body (black, Hispanic, vs. other) and percentage of low-income students (< 80% vs. ≥80%), because our prior study in the same geographic area showed that student MVPA was lower in minority-majority schools and lower income schools [9]. For middle schools only, the models were additionally adjusted for grade (8th vs. 6th or 7th), because bivariate analysis found that grade was a significant factor only in middle schools. For high schools only, the models were additionally adjusted for the number of male students observed (3 or 4 vs. 1 or 2), because bivariate analysis found that the number of male students observed was a significant factor only in middle schools. Based on the mixed models, odds ratios (ORs) and 95% CI were estimated. A significance level was set at 0.05. All analyses were performed using SAS 9.4 (Cary, NC).

This study is one of the largest studies with observation of over 2000 PE classes over time to evaluate the association between environmental characteristics of PE lessons and student PA levels in underserved communities. However, several limitations should be acknowledged. First, student PA levels were measured through SOFIT observations, which could have been subjective. The SOFIT could also have introduced potential error in the evaluation of student activity levels from focal child observation [25]. However, the SOFIT is the most widely used tool to evaluate PA during PE lessons and has been nationally recognized as an appropriate surveillance tool [19]. Second, despite our efforts to achieve consistent SOFIT coding across observers as well as over time within an observer, we failed to record any data to quantify reliability, which could have compromised the quality of the SOFIT data. Third, although we observed over 200 PE lessons in middle schools, these were only based on six schools. Therefore, caution is needed to interpret the middle school results. Fourth, teacher’s participation in the PE teacher training could have affected PE class characteristics and student PA levels. Fifth, despite our efforts to follow the guidelines for SOFIT reporting [11], we did not have data for other environmental conditions (e.g., size of instructional space, available equipment and supplies), teacher characteristics, or the quality of the lesson content, which could have affected student MVPA. Thus, unmeasured confounding could have biased our results. Lastly, the findings from this study may not be generalizable to any other settings or populations.