Skip to main content
. 2019 Feb 1;16(3):429. doi: 10.3390/ijerph16030429

Table 1.

Methodological biases in sampled papers.

Bias Category Biases Identified
Study design

  1. Randomized control trial rather than observational data (4 pts.)

  2. For observational studies, multiple years of data for outcome variable (1 pt.)

  3. For observational studies, individual-level data for outcome variable (1 pt.)
Confounding

  1. Adequate control for confounding variables, specifically socioeconomic status (SES) (2 pt.)

  2. Rationale for selection and inclusion of control variables (little or no rationale = 0 pt., empirical or theoretical rationale = 1 pt., both empirical and theoretical rationale = 2 pt.)
Statistics

  1. Used appropriate statistical analyses for given dataset(s) and research question(s), such as a detailed description of the statistical technique used, explanation why this technique was chosen, and discussion of caveats regarding the conclusions drawn from analyses using this technique [40] (1 pt.)

  2. Performed sensitivity test(s), for instance, differential effects by urbanization, gender, SES, or distances in which green space was measured (1 pt.)

  3. Tested for potential non-linear relationships between green space and outcome, for instance, splitting green space into deciles or tertiles (1 pt.)

  4. Corrected for correlation between variables using a reasonable cut-off value (VIF < 3.0) (1 pt.)

  5. Did not consider pairwise error rates when reporting a large number of analyses, which affect Type I (false positive) error rates [41] (−1 pt.)

  6. For geospatial analyses, did NOT control for spatial autocorrelation, which results in correlated residuals and unreliable model results [42] (−1 pt.)

  7. For multi-year studies, did NOT control for temporal autocorrelation, which also lead to correlated residuals and unreliable model results [42] (−1 pt.)
Exposure assessment (for geospatial studies that rely on large datasets to measure green space)

  1. Multiple seasons of green space data to control for seasonal fluctuations in measurement [43] (1 pt.)

  2. Multiple years of green space data to control for annual fluctuations in climate affecting measurement [44,45] (1 pt.)

  3. High resolution green space data (more than 50 m = 0 pt., 20 m to 50 m = 1 pt., 1 m or less, 2 pt.) to limit under- or over-estimating green space quantity across urban-rural gradients [46]

  4. Green space data not aligned in time with educational outcomes, for example green space data from 2004 and educational outcomes from 2012 [47] (−1 pt.)