# Thread: Need help analyzing data in SPSS

1. ## Need help analyzing data in SPSS

Hello everybody

We conducted a simple experiment, our hypothesis was that negative biometeorological forecast induces nocebo effect, especially in pessimistic people.

We had a control group who watched a regular forecast video and experimental group watched the same video but with negative biometeorological message in the end.

We measured optimism-pessimism with LOT-R (scale variables) and used a part of METEO-Q with a list of psychophysical symptoms (scale as well) to asses their state in past 25 hours.

SPSS showed no significant difference between groups.

Now we want to check if pessimistic participants of experimental group showed more symptoms but we're not sure how to do that. We want to test if the expected effect is shown only in pessimistic people. Any ideas?

2. ## Re: Need help analyzing data in SPSS

Are you using regression of ANOVA to do this model?

I am not sure from your comments if the measure of pessimism is interval, ordinal, or nominal. And the same for your dependent variable (I am not sure exactly what your dependent variable is). Depending on which they are you can run different forms of regression with the predictor if they are optimistic or pessimistic against symptoms. You would code your predictor as either a dummy or interval variable depending on what form the data is in (that is if its interval or not).

You need to provide more details on how you measure your variables. Scale can mean many forms of data some nominal, some ordinal (this is most common) and some interval.

3. ## Re: Need help analyzing data in SPSS

Our dependent variable is the result on meteo-q, a list of symptoms on interval scale from 0 to 3.
Optimism/pessimism - 6 items on interval scale from 0 to 4.

Thank you for your effort, we're miserable

4. ## Re: Need help analyzing data in SPSS

noetsi has a good point. Knowing which analysis you ran will help. If you did an anova, you can include pessimism as another independent factor in the model and determine if it significantly interacts with your grouping variable (control and experimental). However, this interaction will examine if pessimistic participants of the control group also have more or less symptoms.

It seems your dependent variable is 'symptom' as measured by your METEO-Q. Your independent variables of group (control and experimental) is a categorical variable. What I'm unsure on is if your pessimism measure (as measured by the LOT-R) categorize people into either being pessimistic or optimistic or is continuous. If it is a continuous variable you cannot include it in the ANOVA the way I have suggested because it must be a categorical variable and so you would have to go the route of a regression as noetsi has suggested.

5. ## Re: Need help analyzing data in SPSS

Our independent variable is continuous but it could be recoded into dummy variable. We'd like to keep it continuous if there's a way to do it. Is regression analysis a solution?
Also, we'd like to control for general health (interval scale from 1 to 5) and chronic diseases (1 nominal variable -yes/no). How should we do that?

6. ## Re: Need help analyzing data in SPSS

I don't know if ANOVA proper handles interval IV, but ANCOVA certainly will. You don't want normally to recode an interval variable to be a dummy variable.

Regression or ANCOVA (they are essentially the same thing) will deal with interval IV. But while you called your DV interval it appears to only have 3 levels so I am not at all sure you can treat it as an interval variable. In regression DV that have a few levels generate a series of errors such as heteroskedacity unless you use logistic regression.

If your DV has three ordered levels I would 1)do ordered logistic regression or 2) recode the 3 levels into 2 and do binary logistic regression. Be sure the assumptions of ordered logistic regression are met (they vary from unordered or binary logistic regression in some regards).

7. ## Re: Need help analyzing data in SPSS

Oh, sorry, I wasn't clear enough
DV doesn't have 3 levels, it has 15 items on scale from 0 to 3 (e.g. Did you have a headache in the past 24 hours?).
Optimism/pessimism is not IV, our IV is control or experimental condition. Participants were not randomly assigned in conditions, we measured optimism/pessimism after the experimental manipulation in both groups. Therefore, we want to test if pessimistic participants were more subjective to negative biometeorological message to prove that pessimistic individuals are more responsive to nocebo effect.

8. ## Re: Need help analyzing data in SPSS

Or you could do hierarchical/sequential logistic regression. Hierarchical multiple logistic regression is just a variant of basic logistic multiple regression in that you are just specifying a fixed order of entering variables into the model to control for the effects of the covariates (e.g., general health). Then in the next model you place all your predictor variables of interest.

Model 1=variables you want to control for
Model 2= variables you want to control for plus predicator variables.

Comparing the R square change between these two models will allow you to how much more the predicators (model 2) can explain for the variance and reduce your error in predicting the dependent variable of symptoms than model 1 (with just the covariates). By doing this, you'll know how much more variance in the dependent variable are explained by the predicators above and beyond the covariates you are controlling for.

Including the group* pessimism interaction in the model term will help answer your question "if pessimistic participants of experimental group showed more symptoms"

I hope that makes sense.

9. ## Re: Need help analyzing data in SPSS

Regardless of whether your dependent variable is ordinal or interval, I would recommend hierarchical linear or logistic regression. If your dependent variable is an average score across 15 items, you could argue it is continuous and do hierarchical linear regression.

10. ## Re: Need help analyzing data in SPSS

If, and I am not sure this is the case, your dependent variable has 15 different items on a 3 point ordinal scale (like a likert question which is common in surveys) and you combine those 15 items into a single one which you then use as the DV then it is generally agreed that this combined item is interval (or interval enough for regression). In fact in this case since you would have answers from 0 to 45 (45 distinct levels) its doubtful if ordinal logistic regression would work correctly (its not reccomended when you have that many distinct levels).

I am still not entirely sure whether optimism/pessimism is measured (interval, ordinal, or nominal) but however, that is done you could use it as either an interval variable or a dummy variable. It is not clear to me if you are simply arguing that pessimism has an effect on your DV and you need to control for it, or you are arguing that the message (i.e., experimental versus control group since only the former got this message I believe) has a different impact at varying levels of pessimism optimism. If you think pessimism has an independent effect you want to control for then your would have DV=Pessimism + Control/Experimental

If you think control/experimental's impact varies at specific level of pessimism, then I think you would specify an interaction effect and see if its signficant. Your model would be DV= Pessimism + Control/Experimental + Pessimism*Control/Experimental (this is your interaction term). If your interaction term is signficant, then the impact of your groups varies at specific levels of pessimism.

Note I assume in this you measured everyone in all groups for pessimism. Not doing random assignment based on this causes all the normal issues with DOE (causality is weakened since external factors for example could also influence this)

11. ## Re: Need help analyzing data in SPSS

Optimism/pessimism is interval and we're arguing that IV impact varies at specific levels of pessimism. Also, we measured everyone for pessimism.

So we should do a hierarchical regression analysis. Should we center our continuous variable - optimism/pessimism? How to check for interaction in that model using SPSS?

12. ## Re: Need help analyzing data in SPSS

I think you should check for an interaction effect if you believe that the specific level of pessimism influences the difference between your control/experimental group on your dependent variable. Hiearchical regression, as I understand this, tells you if one variable adds predictive power to your model given that another variable is already in it. It does not tell you if the impact of your IV varies at a specific level of another IV - which is what I think you are interested in.

You can specify an interaction term in SPSS by creating a new variable that is (in this case) pessimism*control/experimental. That is the value of one IV times the value of the other IV. I think you can do this under the Transform drop down with Compute Variable but it has been a long time since I worked in SPSS. You would then add this variable to your regression just like any other

13. ## Re: Need help analyzing data in SPSS

Originally Posted by Ywannah
Participants were not randomly assigned in conditions
Why not?? Random assignment isn't always feasible for practical or ethical reasons, but in a study like this it should have been quite feasible. Without random assignment, how do you know that any differences you do observe aren't just due to pre-existing differences between participants?

14. ## Re: Need help analyzing data in SPSS

I got the feelin they did not consider it that important until after the data was gathered. Which happens a lot.

15. ## Re: Need help analyzing data in SPSS

Originally Posted by noetsi
Yeah, in some real-world settings it's very difficult to use random assignment... so you have to think very carefully about confounds and use whatever alternative control strategies you can to try and bolster the trustworthiness of your findings.

But the study we're talking about in this thread is (or seems to be) a laboratory study. That's a completely different kettle of fish. I can't understand why someone would run the study as described without using random assignment. Maybe I'm missing something.

 Tweet

#### Posting Permissions

• You may not post new threads
• You may not post replies
• You may not post attachments
• You may not edit your posts