In our study of digital twin prediction of business software SUS scores, we found high correlation (.80) with human scores but average errors of about +10 points (AI prediction consistently higher than human scores). measuringu.com/digital-twins…

Randomization tests use computer functions to randomly shuffle data many times. To evaluate mean differences after each shuffle, the means computed from the shuffled data are compared with the observed mean difference. measuringu.com/randomization…

The real value is not in the equipment or technology but in the technique. Here are the basics of what you need to moderate a #usability test. measuringu.com/conduct-moder…

When setting #benchmark targets, aim high but realistic. measuringu.com/benchmark-tar…

“Surely, God loves the .06 nearly as much as the .05” (Rosnow & Rosenthal, 1989, p. 1277). measuringu.com/setting-alpha…

For research contexts in which classical test theory provides inadequate measurement of individual differences, use item response theory (including Rasch analysis) instead. measuringu.com/applying-rasc…

A bipolar measure of likelihood to recommend appears to better predict negative word of mouth than the NPS measure of likelihood to recommend, but it cannot be interpreted with current NPS benchmarks. measuringu.com/does-the-nps-…

What is the N−1 two-proportion test, and why do we recommend it over chi-square and Fisher exact tests for analyzing 2 x 2 tables? measuringu.com/what-is-the-n…

Methods for conducting surveys include face-to-face, telephone, mail, and online. measuringu.com/census-poll-s…

Our confirmed version of a new standardized questionnaire for perceived website clutter has separate measures of Overall Clutter, Content Clutter, and Design Clutter. measuringu.com/confirming-a-…

Unlike our previous research, we found no significant differences in predicted SUS scores from two types of digital twins (summary and full)—but both were very different from human scores. measuringu.com/digital-twins…

The SUPR-Q has high internal-consistency reliability (Cronbach-alpha = .86). measuringu.com/10-things-sup…

How Much Should You Over-Recruit? measuringu.com/how-much-shou…

Math and empirical data indicate that the first few users in a formative #usability study will uncover most of the common problems that are available for discovery, but how can this help when planning these types of studies? measuringu.com/predicting-n-…

In an experiment with digital twins predicting SUS scores for business software, we found similar standard deviations for humans and AI but very different shapes (skewness and kurtosis). measuringu.com/digital-twins…

The Product Market Fit item score is the percentage of respondents who would be very disappointed if they could no longer use a product that they have been using. measuringu.com/product-marke…

As a #UX researcher, have you ever worried about using the wrong rating scale format? Would it surprise you to learn that it often doesn’t matter? measuringu.com/changes-to-ra…

User-Centered Design and Design Thinking: Different Origins, Similar Practices. measuringu.com/ucd-and-desig…

Our analyses of three versions of a technical activity checklist indicate that a list of 16 activities is too much, 9 isn’t quite enough, and 10 is just right. measuringu.com/rasch-analysi…

Our study of clothing websites found that they provide experiences that are above average (collectively, at the 82nd percentile in the SUPR-Q). Individually their SUPR-Q percentiles ranged from the 50th percentile (H&M) to the 98th (Banana Republic). measuringu.com/ux-nps-benchm…