Many measurement biases affect student evaluations of instruction (SEIs). However, two have been relatively understudied: halo effects and ceiling/floor effects. This study examined these effects in two ways. To examine the halo effect, using a videotaped lecture, we manipulated specific teacher behaviors to be ''good'' or ''bad'' based on specific items on the evaluation form (the Teacher Behavior Checklist). To examine ceiling/floor effects, we expanded the standard 5-point rating to either 7 or 9 points. Across three universities, 537 students watched and rated the videotaped lecture. Both biases were robust and remained despite characteristics of the measure designed to combat them. Instead, direct instruction regarding how to use the evaluation forms seems to be necessary to limit these biases on ratings.Keywords student evaluation of instruction, halo effect, ceiling/floor effect Student evaluations of instruction (SEIs) have become a commonly used tool by which institutions of higher learning measure their instructors' effectiveness. SEIs can have a great impact, positive or negative, on a professor's career-with promotions and even job security at stake. Therefore, it is imperative that we ensure these

Internal consistency reliability refers to the degree to which separate items on a test or scale relate to each other. All things being equal, tests with higher internal consistency more accurately measure the intended construct of the test developers. The most face valid way to measure internal consistency is to split the test into two equal halves and compare the reliability of each, known as split‐half reliability. The three most commonly used statistical tests for measuring internal consistency are the Spearman–Brown, the Kuder–Richardson 20, and Cronbach's alpha formulas. Cronbach's alpha is the most frequently used because it calculates all possible split half values of the test. Using statistical programs, test developers can also measure changes in Cronbach's alpha if questions were to either be added or removed. This method enables test developers to create a psychometrically sound test without including unnecessary test items.