#### ashleymichelle

##### New Member
Hello all,

Someone please help. I'm conducting a study on how group gender ratios influences stereotype threat within career choices and level of gender identification, as well as how primes that blur intergroup biases can influence these DV's as well.

IV1: Gender ratio (categorical) (3 conditions)
IV2: Prime (categorical) (3 conditions)
o DV1: Career chosen (M/F) (nominal)
o DV2: Change in gender identity (Ordinal) - likert scale

Each participants is exposed to one of the three conditions of IV1 & also exposed IV2, giving me a 3x3 independent variable conditions. However because my DV's are ordinal and nominal I'm unable to use MANOVA tests. What would be an equivalent and what is the process I have to take? It is a between participant study design as is looking at differences between the different conditions.

I also want to see the interaction between the two IV's;

Thank you!

#### Masteras

##### TS Contributor
Each participant is exposed to all three conditions of each IV or just one of them?
Secondly, I think you have to treat each of your DV separately.
Multinomial logistic regression and ordinal logistic regression respectively.

#### ashleymichelle

##### New Member
They are exposed to one condition of each IV,
If the change in gender identification uses a difference in two likert scales, would this therefore be interval data or would it remain ordinal?
Would this be a between-group design?

Thank you so much for the help!

#### Masteras

##### TS Contributor
Certainly, no difference in likert scales makes sense. As for the other, I have to think. Can you explain the experiment with more details, describe everything.

#### ashleymichelle

##### New Member
Each participant is exposed to two IV's (group ratio and type of prime), and randomly assigned one of the 3 conditions within each IV, so there is 9 possible combinations for each participant to be exposed to.
e.g. (IV1) 3 females: 1 male (majority) & (IV2) gender similarities prime

The two DV's will be measured for each participants individually after being exposed to both IV's: change in gender identification (using difference in scoring between 2 likert scale questionnaires) and a choice of gender stereotypcial job role (male/female).

#### Masteras

##### TS Contributor
each participant will be exposed ot only 2 out of 9 combiantions, right?
Secondly, you do this twice for each participant?

#### ashleymichelle

##### New Member
IV1 = 3 conditions
IV2 = 3 conditions
3x3 = 9 possible combinations of IV conditions, so each partipant would be exposed two 1 combination out of 9 (1 condition from IV1 + 1 condition from IV2)

Then after that the 2 DV's are looked at once for each participants, so the job role they chose & the change in gender identification

Thank you so much for this you're a life saver!

#### Masteras

##### TS Contributor
So, to out this way, there are 9 IVs, and each participant has a value for only 1 of them and two values for the 2 DVs, one for each. How many participants do you have?
For each of these 9 combinations, do you have say 5-6 participants?

We are getting there now, I believe

#### ashleymichelle

##### New Member
yes that is a different way of putting it! yes about 5-6 for each

#### Masteras

##### TS Contributor
Now we are good. The DV is likert scale, right?
You have say 9 * 5 = 45 (patients, indepndent patients).

According to these, it looks like ordinal regression, since the DV is in likert scale.
The IV are to be treated as categorical, which means the analysis will not be very powerful, or acurate.

#### ashleymichelle

##### New Member
yes that is right! 1 DV is Likert and the other DV is nominal

So I will carry out 1 test for ordinal regression and the other would be multinomial for the other?
Would I be able to observe the main effect/interaction of the two IVS on each DV's with those?

#### Masteras

##### TS Contributor
I am afraid not, because you already have too many parameters. Putting interaction terms could reult int he saturated model (I believe so). Even if it does not, still you have few measurements and many parameters.