0

We conducted this very simple 2x2 Between Subjects Factorial Design Experiment, where the DV were their answers to a classic insight problem. Since there is only one correct answer to the problem, the values for the DV were either 1 if the answer was correct and 0 if it was not.

The results I got was an F-value of .000 and Sig. (p value) of 1.000 for one variable and the interactio n. Is there something wrong with my data? Perhaps the DV shouldn't have been dichotomous? What do you suggest I do?

Nolia
  • 3
  • 1
  • 2
  • 2
    Are you using ANOVA for binary data? – sitems Oct 10 '12 at 07:29
  • Yes, it turned out that way since the responses (the DV) is either right (1) or wrong (0). Should I have used another test? Or should I change my scoring so it would not be binary? – Nolia Oct 10 '12 at 08:11
  • According to [this nice article](http://oralpathol.dlearn.kmu.edu.tw/case/Journal%20reading-intern-08-12/statistical%20use-review-BJOMFS-2008.pdf), there are also methods how to test differences between groups when we have only nominal variables (in your case binary). So either try to get interval data suitable for ANOVA or try to use any other test. – sitems Oct 10 '12 at 08:21
  • @MiroslavSabo Answer please! I want to upvote (= –  Oct 10 '12 at 08:40

1 Answers1

1

Ok, so either to get interval data or you can use any technique that is applicable when working with nominal variables (since binary is only a special case). In this article, tutorial on how to choose the most appropriate test is available together with examples so you can compare your situation to any from that article.

In summary, Cochran, McNemar and chi-square tests are widely used when testing group differences with nominal instead of at least ordinal variables.

sitems
  • 3,649
  • 1
  • 25
  • 52
  • I solved it again manually by using Chi-Square, the thing is, the observed and expected frequencies all ended up to be equal, which means the chi-square value is zero. I am beginning to seriously worry about the appropriateness of everything. But the data we got was definitely equal across conditions for both variations of one of the variables. – Nolia Oct 10 '12 at 18:45
  • Ok, could you in detail describe the data again? And also your hypothesis? Maybe something is wrong. – sitems Oct 10 '12 at 18:50
  • And attach also table, please. – sitems Oct 10 '12 at 18:51
  • The design is 2 (Close-Other vs Distant-Other) x 2 (Negative Emotion vs Neutral Emotion). Our hypothesis is that respondents under the Close-Other--Neutral Emotion would more likely be able to answer the insight problem, with the Distant-Other--Negative Emotion as least likely to be able to answer the given insight problem. So for the DV, its either they were correct or not. Close Distant Total Neutral 4 4 8 Angry 2 2 4 Total 6 6 12 – Nolia Oct 10 '12 at 19:00
  • sorry, the table. – Nolia Oct 10 '12 at 19:08
  • Chi statistic is zero for such a table, so this is correct. But note that chi-square test can be applied only if expected values are at least 5 in significant part of cells. It is not this case. You have very small dataset, I am not sure whether such a small data can be tested. Also, chi-square test may be wrong choice for such data – sitems Oct 10 '12 at 19:15
  • I see. Perhaps we need to consider redoing everything. Thanks so much for your help. But just to be sure, having a zero chi-square value would mean that there is no association between the variables, am I correct? – Nolia Oct 10 '12 at 19:24
  • Yes, if the test is appropriate for any situation, then small values of statistics indicate that we cannot reject null hypothesis (about no association between variables). Note that we cannot prove null hypothesis, we can only reject or not to reject it. – sitems Oct 10 '12 at 19:41
  • 1
    I have found similar question, maybe it will be [helpfull](http://stats.stackexchange.com/questions/5935/anova-on-binomial-data). – sitems Oct 11 '12 at 11:36