Exam 2 PS 306, Spring 2004 1. Briefly define the term confound. Then, using a very explicit example of practice effects (maybe even with numbers?), illustrate why conducting a repeated measures experiment without counterbalancing makes the study confounded. How does counterbalancing eliminate the confound? [10 pts] A confound is a threat to internal validity in which some systematic but unplanned factor may well explain our results. In general, a confound emerges when we treat our conditions differently in more than one way. That is, the conditions may differ because of the IV, but they also may differ because of another factor. Let’s model a practice effect as a simple additive effect (+5). That means that the second time that a person is tested, that person’s score will improve by 5 simply because of being tested a second time. Below you’ll see the data from 6 participants. Note that they have different initial scores (because of individual differences). However, if we do nothing to them, but simply test them twice, their scores will improve (due to the practice effect) as seen below: Participant Time 1 Time 2 1 4 9 2 8 13 3 3 8 4 7 12 5 5 10 6 6 11 So, that’s what our data would look like if all that happened was practice. But suppose that we have two conditions (Control and Experimental). Let’s model the treatment effect as a simple additive constant (+2), so that the Experimental Condition would go up by 2 and the Control Condition would go up by 0 (no treatment). Adding that information to the data above would yield a data set that looks like this: Participant Control (Time 1) Experimental (Time 2) 1 4 11 2 8 15 3 3 10 4 7 14 5 5 12 6 6 13 There would now be a “large” difference (7) between the two conditions (5.5 for Control and 12.5 for Experimental). However, that difference emerged not because of the treatment difference alone, but also because of the practice effect (and no counterbalancing). If we were to appropriately counterbalance this design (e.g., have Participants 1, 2, and 3 get the Control Condition first and the Experimental Condition second; and haveParticipants 4, 5, and 6 get the Experimental Condition first and the Control Condition second), the data would look very different. For example, consider Participant 4. Participant 4 would have a Treatment Condition score of 9 (initial state of 7 + 2 for the treatment + 0 for practice effect) and a Control Condition score of 12 (initial state of 7 + 0 for the treatment + 5 for practice effect). Participant Control (Time 1) Experimental (Time 2) 1 4 11 2 8 15 3 3 10 4 12 9 5 10 7 6 11 8 Note that now the difference between the Control Condition and the Experimental Condition is more modest. In fact, the difference (2) reflects only the treatment effect. The practice effect is now distributed equally over the two conditions. Note that one side effect of counterbalancing is to increase the variability within each condition (which will serve to increase the error term). 2. OK Jeff, here’s an interesting study. Gangestad, Simpson, Cousins, Garver-Apgar, and Christensen (2004) studied women over the course of their menstrual cycles to determine if they had a preference for male behavioral displays. I’ll reconstruct their study as a two-factor independent groups design, while retaining the basic message of their article. Women watched a videotape of a male being interviewed. Half of the women saw the male respond to a question about himself (“Please tell me about yourself, including who you are, what you like to do, etc.”). The other half of the women watched a videotape in which a male responded to a competitor for a date with a young woman (detailing why she should prefer to go on a date with him). For each video, one-third of the women responded on Day 3 of their menstrual cycle (a low fertility day). One-third of the women responded on Day 11 of their menstrual cycle (a high fertility day). Another third of the women responded on Day 21 of their menstrual cycle (a low fertility day). The dependent variable is a rating (on a 5-pt scale) by the women of the attractiveness of the male on a short-term basis. High scores indicate that the males were judged to be attractive for short-term sexual affairs. Complete the source table below and analyze these data as completely as you can. [15 pts] 2 15.267 7.633 10.477 <.0001 20.954 .9921 10.678 10.678 14.656 .0002 14.656 .9792 9.356 4.678 6.420 .0025 12.841 .90684 61.200 .729DF Sum of Squares Mean Square F-Value P-Value Lambda PowerDay of CycleVideoDay of Cycle * VideoResidualANOVA Table for Attractiveness15 2.800 .862 .22315 2.533 .743 .19215 4.200 .941 .24315 2.600 .828 .21415 2.533 1.060 .27415 2.333 .617 .159Count Mean Std. Dev. Std. Err.Day 03, CompetitiveDay 03, Pers InfoDay 11, CompetitiveDay 11, Pers InfoDay 21, CompetitiveDay 21, Pers InfoMeans Table for AttractivenessEffect: Day of Cycle * VideoThe interaction is significant, F(2,84) = 6.42, MSE = .729, p = .003. Thus, the first step would be to create a graph of the means as seen below. It appears that there is little difference between Type of Video at Day 3 and at Day 21. However, at Day 11 it appears that the Competitive Video yields a higher rating than the Personal Information video. 2.22.42.62.833.23.43.63.844.24.4Cell MeanDay 03 Day 11 Day 21CellPers InfoCompetitiveInteraction Line Plot for AttractivenessEffect: Day of Cycle * Video The next step would be to compute Tukey’s HSD as a post hoc test. ! HSD = 4.14.7315= .91 I can now say that the attractiveness ratings at Day 3 and Day 21 did not differ significantly. However, at Day 11 (a high fertility day) women rated the male in the Competitive video as more attractive (M = 4.2) than the male in the Personal Information video (M = 2.6). 3. In a study by Baron, Burgess, and Kao (1991), male and female participants read accounts of stories that included a description of a sexist act perpetrated by either a male or a female against a female. The 193 participants described the perpetrator in a way that could be scored for intensity of sexist behavior. The displayed sexist behavior was rated 1 for slightly displayed to 7 for extremely displayed. Part of their Results section reads: Perpetrator gender and participant gender main effects were both significant. Female participants, compared with male participants, gave more intense ratings to both male and female perpetrators...: F(1,189) = 5.06, p <