In all of these papers that show a measure of variance, the error bars on the Figures (or the SD reported in the table) are wrong, because they’re calculated for each condition separately, while the analysis is conducted on the differences. You’ll need to make sure students do not replicate this error in their reports, as they will be penalised for doing so.
1. Composite faces
AW: Overall, I thought this was an interesting sub-topic with some good potential for students to develop a new study (don’t forget, we need them to design things here, not just replicate).
AW: I cannot access that panopto lecture from your link. Perhaps consider making the lecture publicly available (e.g. on youtube as CC-BY-SA) to avoid this kind of issue.
Effect size is OK (Exp. 1 partial eta-squared = 0.29 on face type), although you should make students aware that this is based on a small sample size (19 participants) and so may not be good estimates.
AW: I’m dubious about focussing students on Exp. 1 given the conclusion “this experiment has failed to provide evidence of configurational effects”. Exp. 2 in contrast seems quite interesting, in that inversion speeds up the same-different judgement of chimeric faces, partial eta-squared = 0.29. Small sample (N=22). This can be a one factor analysis as same/different top half seems to have no effect anyway at this presentation interval, and they could perhaps just use different-top stimuli? What do you think?
Hugenberg & Corneille (2009)
Demonstration of the composite face effect, plus some social psychology (in-group / out-group). Effect size is OK (partial eta-squared = .28 on face alignment). Sample size OK.
AW: Analysis potentially a bit shonky? – why only analyze accuracy on ‘same’ trials? They say this is common practice, but the two citations seem to be other social psych. papers, so not sure how widespread this practice is, and it seems like a bad idea anyway? What would you advise student here, Chris?
Zhao and Hayward (2010)
Gender judgements harder with aligned than non-aligned composites. Large effect size (partial eta-squared = .47 on face type) but based on a small sample, so you should make students aware that this may not be a good estimate.
2. Image manipulation
AW: Did I get this right, the main theoretical interest here would come in part 2 where you show the size of some of these effects was modulated by e.g. inversion? If so, may be worth saying this in these notes, as otherwise students might perceive the first part as rather obvious, as all they seem to show is that (a) if you degrade a picture it gets harder to recognise (Balas, Binderman), and (b) if a picture is different, it’s harder to say it’s the same person than if it’s the same picture (Estudillo)
Balas et al. (2019)
Blurring faces makes them harder to recognise. Large effect size (exp. 1 partial eta-squared = .47).
Bindemann et al. (2013)
Pixellating images makes them harder to match. Very large effect size (partial eta-squared = .82). Technically, this is based on a bit of a small sample size (N=20) but the risk of non-replication seems low in this case.
Estudillo & Bindermann (2014)
Changing the image viewpoint makes faces harder to recognise and match. Sample size OK (N=40), large effect sizes (recognition partial eta-squared = .31, for matching partial eta-squared = .41).