# Logistic Regression with SPSS

This question was posted the Assessment and Surveillance forum area and has 5 replies. You can also reply via email – be sure to leave the subject unchanged.

### Diallo zakaria

etudiant en securité alimentaire et nutrition

Normal user

24 Jul 2017, 06:34

In a study on the analysis of the causes of malnutrition we used the chi-square test to see the relationship between dependent variables W/H, W/A and H/A and variables such as age, gender and other characteristics of the mother. Is it sufficient to draw conclusions or is logistic regression mandatory? What is the step to make the logistic regression with Spss? Which are the variables to use for logistic regression?

### Blanche Mattern

Normal user

24 Jul 2017, 10:29

Hi!

I would be interested to know a bit more about the study you are currently doing, model used, variables etc.

We are also working on causes of undernutrition analysis but using a mix-method approach so I'll let other colleagues answering your question, but will be happy to exchange on the work you are doing.

Best,

Blanche

### Nakai

MScIH student, Heidelberg University

Normal user

24 Jul 2017, 10:29

I found a simplified answer that could help you here

http://www.theanalysisfactor.com/chi-square-test-vs-logistic-regression-is-a-fancier-test-better

### Jay Berkley

Frequent user

24 Jul 2017, 11:35

Dear Diallo

From your question, I recommend you to use is the chi squared test to describe your malnourished and non-malnourished children. The chi squared is used for a single variable that has categories (e.g. Male/Female or different age groups) to test whether the numbers in each category are different from what is expected by chance. It gives you information on the features of malnourished versus better nourished children.

Logistic regression is used to test the effects of more several variables that may be related with each other and with the outcome (this is called confounding). However, to get correct results requires more understanding than just how to run the command. You would need to work with a statistician. For a pre-planned study, it would be usual for a statistician to be involved in the design. For analysis of existing survey data, a statistician will be able to interpret the relationships between variables, decide which to include and the see the limitations of the design.

It is also important note that these tests will tell you what features are associated with malnutrition, that does not mean they are the cause.

### Tchapya

Normal user

24 Jul 2017, 14:42

Dear Zakarie

I would recommend using regression logistic as the variable you want to explain is malnutrition. This variable is qualitative and binary (are malnourish /are not malnourish). Other variables such as age, gender and mother social characteristics are explanatory variables. I don’t know how it works with SPSS, but I can help if you can use R software.

Regards,

### Bradley A. Woodruff

Self-employed

Technical expert

24 Jul 2017, 15:17

Dear Diallo:

I will echo Jay Berkley's response. If you don't understand what a specific statistical test means or how to use it, you will need to consult a statistician. It is much, much more complicated than just knowing how to write the commands in whatever statistics computer program you are using.

Logistic regression is never mandatory. The analysis method you choose depends entirely on what you want to learn from your data, and determining this requires in-depth knowledge of the subject being studied and the hypothesized causal chains, confounding, and effect modification. Moreover, logistic regression may not be the best multi-variate modeling technique for data from nutrition surveys or studies. It gives you odds ratios, and odds ratios overestimate the risk ratio when the rare disease assumption is violated. In many nutrition studies, especially when the outcome is stunting, this assumption is grossly violated.

So in short, consult an expert. There are many pitfalls and mistakes which can lead an inexperienced person to incorrect conclusions.