Accounting information systems (AIS) research data may suffer from severe non-normality, which, if not handled properly, may lead to incorrect statistical inferences. To address this problem, we empirically evaluate the relative merits of a Two-Step normality transformation proposed by Templeton (2011) compared to four alternative distributions available to researchers (random-normal, original, natural log transformed, and winsorization transformed). Using 45 corporate financial performance ratios (CFP), we investigated three perspectives on measurement validity: construct validity, reliability, and difference testing. We then examined the efficacy of the Two-Step method in the context of business value of IT research—we regressed four IT investment and three control variables on 31 of theoretically relevant CFP indicators. The preponderance of our evidence shows that the Two-Step method consistently outperforms the prominently used alternatives in achieving statistical normality, retaining original series means and standard deviations, exhibiting validity and reliability, and theory testing. Our findings strongly suggest that AIS researchers consider adopting the Two-Step normality transformation when utilizing non-normally distributed data to obtain a more accurate understanding and interpretation of results.

You do not currently have access to this content.