# Thread: Chi-quare and logistic regression with three level independent variable

1. ## Chi-quare and logistic regression with three level independent variable

Dear Friends,

I am currently working on my master's Dissertation and have some problems in finding out how I should do my chi-square-test and logistic regression. Unfortunately my Prof cannot help me so I really hope you could share with me your opinion on the following problem:

- I have a three-level independent variable that represents my three experimental conditions: 1) control condition (no manipulation), 2) weight manipulation A, and 3) weight manipulation B

- and I have a dichotomous dependent variable (promise) where people answered either keep(=1) or break(=0)

Problem 1:
I wanted to test for significance using a chi-square test. However I am confused whether I need to include the control condition in the crosstab or not.

Since the Hypothesis is "There is an association between Weight and Promise." I thought it would ake no sense to include the condition in which no weight was manipulated.

The chi-square tests that includes all three conditions of the Independent variable is not significant (however significant residuals in the cells of both of the experimental conditions).
However the chi-square test without the control condition would be significant at p<.05.

What do you think which chi-square test is the one most logical?

Problem 2:
I have done a logistic Regression to find out about the odds ratio. Therefore I created two dummy coded variables that represent the two experimental conditions:

DC_1 (0,1,0) where 0=control and Manipulation B condition
DC_2 (0,0,1) where 0=control and Manipulation A condition

Do you think it can be justified to do a chi-square that leaves out the control condition and then do, subsequently, a logistic regression that includes the control condition?

Thank you very much in advance! I appreciate any helping idea!

2. ## Re: Chi-quare and logistic regression with three level independent variable

- I have a three-level independent variable that represents my three experimental conditions
I guess that you have three independent groups? Or is this a repeated-measures design?
Since the Hypothesis is "There is an association between Weight and Promise." I thought it would ake no sense to include the condition in which no weight was manipulated.
It is a bit difficult to comment on this, since you didn't explain what
the topic and the research question are, and what you mean by control
or by weight manipulation. And, by the way, you should always state
The chi-square tests that includes all three conditions of the Independent variable is not significant (however significant residuals in the cells of both of the experimental conditions).
There is no such thing as significant residuals in the cells, AFAIK?
There are some rules of thumb for adjusted residuals, but this isn't
the same as significance testing.
What do you think which chi-square test is the one most logical?
The Chi² test with all the groups. There must have been some
reason to include the control group? Moreover, changing the analysis
because the initial results weren't so welcome is considered
inappropriate. Would you have asked the same questions if the
two-group comparison had resulted in p > 0.05 and the 3-group-
comparison had resulted in p < 0.05?

Just my 2pence

K.

3. ## The Following User Says Thank You to Karabiner For This Useful Post:

Ricarda_Rau (06-26-2015)

4. ## Re: Chi-quare and logistic regression with three level independent variable

Dear Karabiner,
Yes I would have asked the same even with other p-values since I am completely confused at the moment. I really don't want to exclude not welcoming results, I just want to do it right :-)

It is correct that I have three independent groups (each n=60). My control group is a group where I didn't manipulated anything. So should work as a reference group for my two experimental groups.

Referring to why I included my control group in the first place is simply because I thought you should include all conditions.
However now, that I am thinking about it logically I don't see the point of including a condition in which I didn't manipulated the feelings of weight, when the question I am trying to answer is: "Is there an association between feelings of weight and Promise (the Y/variable)".

Does this make sense?

Thanks again in advance!

 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