Due to a highly skewed distribution, the conduct problem score was natural logarithmically transformed to obtain a closer to normal distribution. All factors were scaled such that high values indicated poor adjustment and standardized in order to show effect sizes in terms of fractions of standard deviations.
