Hello everyone,
I really need your help!
I am doing my bachelor thesis and I am struggling with the following. I want to do moderation analysis with the following variables:
- IV: Collaboration types (4 different binary variables)
- MV: Obstacles at Innovation (9 different binary variables)
- DV: Total Environmental Innovation (1 continuous scale variable)
I am doing my analysis with SPSS.
Now, I have two questions:
Is the two-way ANOVA (Univariate GLM) the right analysis design I should use?
And, I looked at the normality assumption of the DV and found that it is non-normal, since a lot of firms in my data set actually did not participate in any environmental innovation. Can I still use the variable? Or how should I transform it? Or should I then maybe use a totally different analysis?
Please help me out! Thank you SO much in advance.
Best, Denise
1) No, it's not the correct one because two-ways ANOVA is for two factors and you have much more
2) DV has no need to be normal considered at a whole. In fact, it it was, that would be a sign that IV has no effect on it.
What has to be normal is DV conditionned on the predictors — or, said differently, the errors of the underlying analysis model.
Hi, it would more interesting if you could transform your continuous dependent variable in binary, and then you could apply the multiple logistic regression and to find out what independent variables are influencing the DV, is that, what's the predictors...
I hope helped you
Best regards
@ Douglas: that makes a lot of information lost and adds several arbitrary choices (how many classes [you say « binary » but also « multiple logistic regression »...], and where to set their limits), so should be considered as a last chance method...
Besides, logistic regression is a non-linear model and checking assumptions is more difficult than the linear model underlying analysis of variance...
Thank you @Emmanuel and @Douglas!! Already helped me out a ton! At least I am not totally on the wrong track...
So, what got me still confused is that my statistics book tells me that I need to satisfy the following assumptions for a ANOVA:
1. Independence (given from my dataset)
2. Normality
3. Homogeneity
So, for what part of my data do I need to check what assumptions.
Because for my binary IV and MV I cannot check normality either and if the data is homogenous then I would not really get any significant results, right? And for the DV, as you just told me, I also don't need to check for normality, would I need to check for homogeneity there? And if yes, how would I be able to check it?
THANKS in advance!1
For IV and MV, there is no assumption, so nothing to check more than all usual sanity checks (no incorrect values, at least two different values, and so on).
I'm not sure what is understood by ·« Homogeneity », I guess it is « homogeneity of the variances », that is « homoskedasticity ».
Whatever it is, all assumptions should be checked on the residuals of the model, not on the DV directly.
Okay, so I decided to do a two-way ANCOVA because I believe that I can then see the influences of my moderation best.
According to Leppink (2018; Article Analysis of covariance (ANCOVA) vs. moderated regression (MO...
) I, therefore, need to have independent residuals that have the same variance (homoscedasticity), and that are normally distributedSo I created standardised residuals using SPSS (GLM -> Univariate -> add DV and IV -> Model: main effects of each IV -> Save: standardised residuals -> Options: homogeneity tests).
Now, I the Levene's Test of equality of error variances is significant at .000. Thus, the null hypothesis of equal variances is rejected and I can assume that there is a differences between variances in the population. I also checked the normality assumption in SPSS (Descriptive statistics -> Explore-> dependent list: Standardised residual of DV -> Plots: normality plots with test). That gave me a descriptive table with: skewness of 0.033 ( thus, does not the Leven's test of homogeneity (like I did it) make no sense in my case?
2. How else should I check for these assumptions? How should I make the residuals?
You cannot do ANCOVA if all your IVs/MVs are binary, ANCOVA is when at least one is continuous.
Your results suggests that assumptions do not hold, so you must now search for data transformations or alternative analysis methods.
At this stage, you'd better search for a local statistician to help you...
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS