I'm trying to perform two-way ANOVA to test the significance of an interaction between treatment (raw/ cooked) and proportion (11 proportions) on moisture content. I used SPSS to generate the results. the anova table suggests that the interaction between treatment and proportion is significant p
Christian wrote, "... the plot clearly shows that the lines are parallel with each other."
How can you say that when the Cooked-Raw difference is much larger at P2 than for any of the other proportions? The lines are reasonably close to parallel if you exclude P2, but certainly not if you include it.
Christian wrote, "... the plot clearly shows that the lines are parallel with each other."
How can you say that when the Cooked-Raw difference is much larger at P2 than for any of the other proportions? The lines are reasonably close to parallel if you exclude P2, but certainly not if you include it.
As Bruce mentioned, the interaction is quite obvious: the treatment effect is much bigger at P2 than it is at other proportions.
Why would you think there is not an interaction? Perhaps you are mistakenly thinking that "interaction" can only mean a crossover interaction (where the lines cross). This is not the case; an interaction is any time when the effect of one variable depends on the other variable. A good rule of thumb is: if you can't fully explain the effect of one variable without also mentioning the other variable, then you have an interaction. (In this case the explanation would be something like, "cooked was more moist than raw, especially in P2", or alternatively, "P2 was more moist than other proportions, especially when cooked.)
115
This looks to me like something is seriously wrong in the data. If the proportions are in ascending order, why would P2 be so different? And why would a tiny difference in raw moisture translate into a huge difference in cooked?
My index of suspicion is right up at this point. Unless you have a good underlying scientific explanation for this finding, I'd be slow to believe it.
I agree with all the previous answers: with the definition of interaction and especially on the need to check the raw data (P2 especially).
I agree with Renán and Roberto, P2 must have some problem, outlier, perhaps. There are an inconsistency, too, between the graphs and the analysis. Symbolizing the proportion as a number you may have a continuous variable, better represented by a line and if you have an nominal scale, you may better represent by a bar graph. If it is a continuous efect, you may try to treat the proportion as a covariate and the treatment as a regular effect (nominal scale) in ANOVA.
Hi All,
Thank you for your great responses. Unfortunately, there was an error in the data encoding. Upon re-checking we have identified the mistake. @Stephen, yes, its unfortunate that I always tend to associate "interaction" with lines that intersect. thanks for the enlightenment.
Because of scale.
- Y=a0+a1X+a2X2+a3X1X2(interaction item)+e
- Interaction item should be nominal scale
Or you had better use moderating regression analysis
and have to consider mean centering
cause by multicollinearity.
Hello there, just in case someone is interested in SPSS 23, it is now available in cloud with 30-day free trial and great price after. take a look at https://www.apponfly.com/en/ibm-spss-statistics-base?KAI
There are existed main and interaction effect but you have to transfer interaction item from interval or ratio scale to nominal scale like 2 by 2, 3 by 3 and so on.
also, you have to know sample size of one group will be over 30 because of nomal distribution.
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS