The scenario is follows: Two groups of participants are asked to recall the stimuli presented to them. Group A are presented with both types of stimuli (imagery) on a tablet whereas Group B are presented with both types of stimuli on a computer. Two types of devices (computer and tablet), and two types of imagery (black and white and colorful) are used in this experimental design. I want to measure the dependent variable called "Recall." Screen size is ignored.

Here is my experimental design: To conduct a one-way within subject test ANOVA for the (i) computer (Imagery with two levels black and white and colorful) and another one-way within subject test ANOVA for (ii) tablet ((Imagery with two levels black and white and colorful).

Kindly advise if conducting a single analysis - a mixed-ANOVA (one in-between and one-within subjects test) will be more appropriate than conducting two separate analyses.

  • $\begingroup$ How do you measure recall ? $\endgroup$ – Subhash C. Davar Mar 24 '17 at 13:17

The two-way with one between and one within is better. You might want to consider stimuli as a random variable so you can generalize beyond the stimuli used in the experiments.

| cite | improve this answer | |
  • $\begingroup$ If there is an interaction between the two variables in a mixed-ANOVA "device types" and "imagery," what does that essentially mean? Do I need to follow up with a Post-hoc? Thanks $\endgroup$ – Vyas Mar 19 '17 at 22:41
  • $\begingroup$ The effect of color depends on device. It should be stated in terms of the relative sizes of the color effect, not just that they're different. $\endgroup$ – David Lane Mar 19 '17 at 22:48
  • $\begingroup$ I ran the mixed-ANOVA. For Levene's Test, I obtained one significant value (p<0.05) for B/W and a non-significant value for colorful imagery (p>0.05). I understand that if p>0.05, we are not violating the homogeneity assumption. However, what action should I take as the other level of the factor imagery is significant. Thanks $\endgroup$ – Vyas Mar 19 '17 at 23:14
  • 1
    $\begingroup$ Assumptions like homogeneity of variance and normality are always violated, the question is by how much. ANOVA is very robust to violating the assumption of normality but if you have highly skewed distributions a transformation such as log could increase power. You could use the Welch test if you are concerned with homogeneity of variance. $\endgroup$ – David Lane Mar 20 '17 at 3:28

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.