Subjective sensitivity data: Considerations to treat heteroscedasticity

In many clinical disciplines, especially in subjective sensitivity research, data are collected on the ratio scale. This is also true for vibration perception thresholds (VPTs). Hence, such data may be of heteroscedastic nature, where the amount of measurement error increases as the values increase. Heteroscedastic data is a great scientific concern, since many standard inferential statistical tests assume homoscedasticity and may, therefore, lead to incorrect interpretations. Unfortunately, only few studies even address heteroscedasticity, and even if they do, in many cases this is achieved inappropriately. Therefore, in this paper, we first show a simple approach for how to detect heteroscedasticity in VPTs measured at the plantar aspect of the hallux. Second, we show how to correct for heteroscedasticity by means of logarithmizing the raw data. Finally, we also demonstrate the importance of how data benefits from this transformation. Hence, if data takes benefit from the transformation, we recommend using logarithmized data to conduct further statistical approaches. We also interpret the results presented here. Furthermore, we briefly mention alternatives in case the transformation is not successful in eliminating heteroscedasticity.