Robustness of the t test applied to data distorted from normality by floor effects

Area of Science:

• Oral health clinical trials
• Statistical analysis in dentistry
• Biostatistics

Background:

• Dental trials for plaque and gingivitis often involve pre- and post-treatment measurements.
• Measures like Volpe-Manhold and Löe and Silness scales can exhibit non-normal distributions with a proportion of zero scores, especially in short-term studies or after oral prophylaxis.
• The two-independent-sample t test is commonly used to compare treatment effects.

Purpose of the Study:

• To investigate the robustness of the two-independent-sample t test when applied to dental outcome measures with distributions distorted from normality.
• To assess the reliability of t tests in clinical trials with a significant proportion of zero scores.

Main Methods:

• Computer simulations were employed to model outcome scores from dental plaque and gingivitis trials.
• The study simulated distributions that were approximately normal above zero but included a proportion of subjects with zero scores.
• The performance of the two-independent-sample t test was evaluated using both raw outcome scores and differences between baseline and outcome scores.

Main Results:

• The two-independent-sample t tests demonstrated robustness even with non-normally distributed data.
• Actual significance levels closely approximated nominal significance levels across various conditions.
• Robustness was maintained even with small sample sizes and up to 50% of subjects achieving zero scores.

Conclusions:

• The two-independent-sample t test is a reliable statistical method for analyzing data from dental plaque and gingivitis trials.
• This statistical approach is effective even when dealing with common data distortions, such as zero-inflated and skewed distributions.
• The findings support the continued use of t tests in clinical dental research, ensuring valid comparisons of treatment efficacy.