Do not try to force a specific statistical model on your data set. It is the other way around; the experiment dictates how to analyze the data.

What are the independent variables X's in your study? I will assume that they are all continuous. Given what we know about the dependent variable (Y), I suggest these possible way for modeling:

1. Do a Chi square test only on the dependent variable Y. This will give you some information on the distribution of the values of Y.

2. Do the multivariate analysis of variance F test where Y is the treatment factor and where the X;s are the response variables here. This will be the main statistical analysis that you need. Test for normality first. 

 3. Do a logistic regression with 0=no player 1=otherwise.

4. Do a correlation analysis on the X's. 

Use all of the above to learn about the experiment. Each part gives you some information. 

Rather than selecting

    Analyze > Regression > Binary Logistic

from the menu, select

    Analyze > Regression > Multinomial Logistic

Than you for the answer.

I´ll try.

The creation  of dummy variable can be also an alternative?

Than you for the answer.

However I still have some doubts :)

The creation of dummy variables or reclassify the dependent variable as ordinal instead of nominal and use a linear regression could be alternatives?

Best regards,

Catarina

I´ve tried and i got this message:

The number of dimensions requested for the Observed and Predicted Frequencies table has exceeded the limit (20) of the pivot table subsystem. Therefore, this table will not be produced.

There are 1482 (92,9%) cells (i.e., dependent variable levels by subpopulations) with zero frequencies.

Insufficient memory to complete the model estimation step.

This command is not executed.

Am i doing somthing wrong?

Best regards,

catarina

Anyone knows what the above message means in SPSS output?

(Multinomial Logistic)

warning

The number of dimensions requested for the Observed and Predicted Frequencies table has exceeded the limit (20) of the pivot table subsystem. Therefore, this table will not be produced.

There are 1482 (92,9%) cells (i.e., dependent variable levels by subpopulations) with zero frequencies.

There is possibly a quasi-complete separation in the data. Either the maximum likelihood estimates do not exist or some parameter estimates are infinite.

The NOMREG procedure continues despite the above warning(s). Subsequent results shown are based on the last iteration. Validity of the model fit is uncertain.

Thank you

The problem relates to a high number of missing values in some independent variables.

Is there any way to perform the analysis excluding missing values ?

By default, SPSS does "complete-case-analysis". I.e., it excludes any case (row in the data file) for which any of the variables are missing. You may wish to take a look at using Multiple Imputation--note that NOMREG does support pooling of the estimates (see the link below). If this is not something you have ever done, I suggest that you find someone local to help you. It will become too involved to be solved easily by e-mail or discussion list posts.

I don't think you've said how many explanatory variables you have.  Another issue you need to be thinking about is the possibility of over-fitting.  See Mike Babyak's nice article (link below) for an overview of that topic.

HTH.

http://pic.dhe.ibm.com/infocenter/spssstat/v22r0m0/index.jsp?topic=%2Fcom.ibm.spss.statistics.help%2Fspss%2Fmva%2Fmi_analysis.htm

http://people.duke.edu/~mababyak/papers/babyakregression.pdf

Thank you very much for the answers.

I´ll try to find someone local to help me in this analysis.

Best regards,

Catarina

if you have a response that is at 5 discrete levels, you should use ordinal logistic regression. 

The types of input variables shouldn't matter much. If you have missing data, you can do a few things. Most software will remove those data points. (If you have 100 sets of data and 12 have missing data, you end up using 88 sets of data.) 

You can try to see what the missing data is and find it. (i.e. a patient has a missing birthdate but does have 'age' or vice versa) ( Patient seen at "womens only" clinic is missing gender) 

You can try to predict the missing value or fill the missing value with the mean/median of your data. (I don't recommend this. I have seen it done.)

Your response variable seems to be ordinal. In this case the multinomial logit model is not appropriate. You have to use an adjacent, a cumulative or a sequential model to take account of the ordering assumption among categories. Since the response seems like a process I would prefer sequential models.

For the warning, it means that you have too much explanatory variables (especially if they are qualitative) for a small number of observation. You have to select a subset of explanatory variables.  You can also use proportional models (i.e. the same slope parameter for all categories) which are more parcimonious than the multinomial logit model.

I am in doubt that you can speak of an ordinal scale regarding the DV - this is only the case for the categories 2 to 5, but category 1 (no player) is rather the dichotomous alternative to the other categories. So I think the multinomial model would be appropriate here.

It is true that "category 1 (no player) is rather the dichotomous alternative to the other categories". That's why you can use a sequential model (sequence of dichotomous alternatives)

Do not try to force a specific statistical model on your data set. It is the other way around; the experiment dictates how to analyze the data.

What are the independent variables X's in your study? I will assume that they are all continuous. Given what we know about the dependent variable (Y), I suggest these possible way for modeling:

1. Do a Chi square test only on the dependent variable Y. This will give you some information on the distribution of the values of Y.

2. Do the multivariate analysis of variance F test where Y is the treatment factor and where the X;s are the response variables here. This will be the main statistical analysis that you need. Test for normality first. 

 3. Do a logistic regression with 0=no player 1=otherwise.

4. Do a correlation analysis on the X's. 

Use all of the above to learn about the experiment. Each part gives you some information. 

I think the reason for the non-execution of this command is simple. You chose the wrong analysis (i.e. binary logistic regression) for your dependent variable which had more than two categories. But, regarding whether or not your dependent variable is sequential or not you should know since you did the coding. If it is ordinal then I suggest you use the ordinal logistic regression model but it it is nominal, then use the multinomial regression model.  

The obvious answer is to do multinomial logistic regression (or ordinal logistic regression if the five categories are ordered). A question that comes to mind is: what do you want to use the model for? Do you want to predict whether a player is a risk player (for example) given the values of the independent variables? On the on the hand, if you want to characterise the different categories of players, you might try a CHAID analysis.

It seems that an ordered logit model (also called ordered logistic regression or proportional odds model), is a model for an ordinal dependent variables like the one that you seem to have here. For example, if in your survey you clearly define the categories according to how often the subject played a particular game on a scale that goes from "Never plays (1)", "social player (2)", ...., "clinical pathological player (5)", and the purpose of the analysis is to see how well that response can be predicted by the responses to other questions, some of which may be quantitative, then ordered logistic regression may (should?) be used. All of the statistical packages have this capability built in and since you have worked and have experience with logistic regression in the past, it should be relatively easy for you to perform this analysis.

Stay away from ANOVA and its related F tests because ANOVA requires that your model residuals are (reasonably) normally distributed - which in your case is impossible because you do not have a continuous response variable. As usual, it is not a bad idea to consult your local friendly statistician, specially, if you need guidance on the model assumptions and with the interpretation of the outputs.

As an experienced statistical consultant, I cannot see how solid advice can be given without knowing details about the experiment and knowing what the researcher wants to accomplish.  There should be no guessing. 

Binary logistic regression can only be used if the total number of variables are 2. If the total number of variables are more than 2 as it is in your case then you should rather use multivariable or multinomial logistic regression analysis. Kindly see the link below. 

http://www.unt.edu/rss/class/Jon/SPSS_SC/Module9/M9_MultiNomReg/SPSS_M9_MultiNomReg.htm

I agree with dear Raid Amin.

The outcome variable is not ordinal as in ranking, for example. It is composed of 5 categories that are independent from each other. So multinomial logistic regression should be used. Check this link:

http://www.ats.ucla.edu/stat/spss/output/mlogit.htm

THe multinomial gives errors, "insufficient memory to complete the model estimation step... execution of this command stops" even though the factor variables are complete and all the covariate variables as well. I have tried to run the variables separately to the dependent variable but only one of the variables works, does anybody have experience with Nom Regressions?

Thanks

Rather, I would suggest Econometric software tools such as LimDEP(NLogit) for nominal regressions

In this specific case, the problem is having more than TWO values for your DV, please check your datasheet carefully, and remove/manage those missing or even blank cells