0

On a mixed model on SPSS I got the following results:

in the table called "Type III Tests of Fixed Effectsa"

Source    Numerator df      Denominator df  F        Sig.
...
Genotype    5               96.287          3.542   **.006**

In the table called "Estimates of Fixed Effectsb"

Parameter   Estimate    Std. Error  df  t           Sig.    95% Confidence Interval 
                        Lower Bound Upper Bound
[Genotype=1]    -.378349    .205116 96.779  -1.845  **.068**    -.785459    .028761
[Genotype=2]    -.203366    .283755 94.269  -.717   **.475**    -.766747    .360015
[Genotype=3]    .010477  .211701    99.681  .049    **.961**    -.409548    .430502
...

(The p values are in bold, couldn't paste the table from excel). I'm still learning statistics so I'm not sure if I'm not missing something: what the results mean is the "genotype" is significant for the model test parameter (because the P was 0.06), but all the sub-groups under "genotype" (this is a parameter with 6 different groups) have p values more than 0.05 and that means there is no significant difference between them, so how can the p in the first table be < 0.05??

Tim
  • 108,699
  • 20
  • 212
  • 390
Emili
  • 1
  • 1
    Welcome to CV! Notice that you can use `{}` button to paste a code with more appropriate formatting (I already fixed that for you). – Tim Dec 22 '14 at 08:43
  • While I don't use SPSS I am unsure if I understand the names of the tables correctly, but in general lower-level and higher-level effects are not the same and should not be confused (check for an example [here](http://stats.stackexchange.com/questions/125683/pearson-correlation-has-quizzy-results/125686)). – Tim Dec 22 '14 at 08:47
  • thanks for the formatting and the answer, I can see why the results can be that way, but I still don't understand the practical mean of the results in real life. The model was supposed to find which variables had effect on the result of some dependent variable. What I understand from the first table is that genotype is significant, but from the second table I get there is no significant difference between the groups. In real life each person must be in one of the groups, so can I say there is no real difference between the sub groups? if that so, what is the meaning of the first table??? – Emili Dec 24 '14 at 08:28

1 Answers1

1

Generally when reviewing the parameters estimates table SPSS will choose your last level as the reference with which to compare the other levels of the factor.

To make sense of your case, I assume there is (at least) another line you haven't copied, and that this has redundant estimates (all zero). This seems likely given that the numerator you give for Genotype is 5, which tells me there are probably three more lines you haven't copied [Genotype=4], [Genotype=5], and [Genotype=6] the 6th is the one all the others are compared to (and therefore 0 in the lower table as it is being compared to itself).

If this is the case, then the p-values > 0.05 for Genotype 1, 2 and 3 are in direct comparison to Genotype level 6, and are not significant. However, that says nothing about whether Genotype 1 and 2 are different from each other and this would have to be compared directly.

There are some options to compare the main effects in SPSS when you click the EM MEANS option on the right-hand side, than move Genotype to the box and leave the reference category as "none" which will compare all levels to each other.

Hope that helps!

Mensen
  • 356
  • 1
  • 6
  • (+1) This a good analysis and ought to be helpful for the O.P. Note, however, that the question attests that "all the sub-groups ... have p values more than 0.05". Presumably this means all six, not just the three shown in the question. On this presumption I have pointed this thread at a pre-existing one that asks the same question. – whuber Jan 02 '15 at 15:52
  • Mensen, thanks. As you guessed, there are 6 groups for "Genotype". by what I did not understand was how all of them were with p value > 0.05 while the overall was p<0.05. When I ran this test but change the groups order (i.e.: I recoded 1 to be 7) there was difference between some of the groups. What I wanted to know was how to intercept the results. I guess the results mean the "genotype" as a variable is significant for the paramter the model tested and the difference between groups should be calculated alone. Thanks! – Emili Jan 03 '15 at 17:41
  • @Emili In case you didn't read that far in my post... if you go to the *EM MEANS* box when you are running your mixed model, you can specify 'post-hoc' comparisons for the factor *Genotype* so that all the pairwise comparisons are run for you. – Mensen Jan 03 '15 at 17:47