Random responses inflate statistical estimates in heavily skewed addictions data

Area of Science:

• Psychology
• Statistics
• Addiction Research

Background:

• Random responding in self-report surveys is a concern, particularly in addiction research.
• Screening for random responses is rarely reported in the published addictions literature.
• Random responses can bias statistical estimates, especially in skewed data typical of substance use outcomes.

Purpose of the Study:

• To illustrate how random responses bias statistical estimates using simulated and real data.
• To examine the impact of varying amounts and types of random responses on statistical estimates.
• To highlight the problematic nature of random responses in skewed data common in addiction research.

Main Methods:

• Tested effects of random responses on covariance-based statistical estimates in distributions with varying skew.
• Used simulated data to model random response effects.
• Replicated findings in a real dataset (Add Health) by introducing simulated random responses.

Main Results:

• Skew and the proportion of random responses influenced bias.
• Uniformly random responses deflated estimates in non-skewed data; long-string responses inflated them.
• In skewed data, all types of random responses inflated estimates, even at low proportions (≥2.5%).

Conclusions:

• Failure to screen for random responses leads to biased statistical estimates.
• Even small proportions of random responses (2.5%) can inflate estimates in skewed data.
• Screening for random responses significantly enhances data quality, reliability, and validity.