correlation analysis, i think

Thanks Ayyaz........but I'm looking for association not correlation

association asking for prediction and uses regression tests......as the following:

all regression tests are for single outcome VS predictor.......if:

(single) dichotomous outcome VS single predictor: single linear regression

(single) dichotomous outcome VS multiple predictors: multiple logestic regression

(single) polytomous outcome VS single or multiple predictor: multinumial regression

===> My case Is: (multiple) dichotomous outcomes VS single polytomous predictor ????

I second Ayyaz to use Crosstable test unless you can come up with some other reasonable independent variables (age, gender, environment, etc) otherwise it is kind of hard to imagine other statistical analyese can help you.

hmmmmmm thanks Ru-Jeng

I'm also thinking about it ....  contingency table analysis 

thanks all

I suggest the multiple outcome to be coded as 1 sepsis 2 pneumonia 3 abscess 4 etc...

and the independent variable is bacterial genotypes. You have two choices:

1. Chi square test / fisher exact test: the association between bacterial genotype and respiratory diseases. but this is a univariable analysis.

2. Multinomial logistic regression: you can have multiple independent variables [dichotomous or numerical] and single dependent variable [nominal variable]. This is a more powerful statistical analysis because it is multivariable which confounders can be controlled or adjusted.

best wishes

Thanks prof Norsa'adah

But in my case, the patient may have many outcomes at the same time, not only one.....

for example: a patient with genotype A (the independent) suffered from soft tissue abscess, spread to lungs and caused pneumonia and complicated sepsis 

another patient with genotype B suffered from spleen abscess spreaded to liver, sepsis and septic shock that was fatal. 

so I cannot arrange them as polytomous categories in a single dependent variable.

I'll chose your first suggestion, Chi square test / fisher exact test: the association between bacterial genotype and every outcome singly .... I'll chose the most frequent clinical pictures (outcomes) and do it as univariable analysis for each

thanks prof and thanks in advanced for any comment........

Why are you analysing multiple overlapping outcomes simultaneously? It seems to me that it's close to impossible to describe your dependent variable. If you analyse them separately, then you need to define your comparison group carefully. There is no point in comparing, say, pneumonia with all other patients. The comparator group is then ill-defined. 

It seems to me that you have to step back and formulate your research question in words before you proceed. Does bacterial genotype distinguish between…? Something like that. 

One exploratory method would be to heat map the results in a table. But forget a Chi-squared test. It tests a hypothesis that is too vague to be useful. 

Thanks all................

Hi,

Your question: What is the proper statistical test for several outcomes VS single predictor? 

Best you would prepare a 2 x 2 table, and you can perform a Chi-Sqaure test. this test you should use when all the cell values > 5.  If one cell value < 5, then you can go to Fishers' exact test.  If any one cell value 0 then go to Yates' Chi-Square test.

other wise, apply multiple logistic regression analysis for your study. 

If you have multiple independent variables are in dichotomous or numerical values and single dependent variable are in nominal, then you go for Multinomial Logistic Regression analysis.

All the best.

What is the relationship among you binary variables? Have you tried IRT on them see if they might be summarized by a small number of traits. 

IRT is closely related to a multilevel logistic regression where a set of dummy variables on the right side of the equation say which of these variables is being referred to. Multilevel (also called mixed, hierarchical, etc.) logistic regression would be a procedure that I would consider. The multinomial procedures listed in other responses will be problematic since your respondents could have said yes to multiple items.

Apart from logistic regression test for association and determine the odds of association , I suggest structural modelling for the relationship paying attention to the causal path( i.e the root taken to the dependent (Outcome) from the predictor (independent)

If the output independent of each other, you can apply Chi-square test between the independent variable and all supported variable (independently).

Or logistic regression.

Also you can apply correlation coefficient Point Bacyrial

Hi

1. The sample sizes of the random variables are not given and no information regarding the relation between the random variables is provided. Several tests could be developed.

2. Please refer to:

1. Landau, S. and Everitt, B. S.(2004). A Handbook of Statistical analyses using SPSS.

2.Howitt, D. and Cramer, D.(2008). Introduction to SPSS.

Regards,

Zuhair

following

interested

interested

if your outcome variables and also predictors are quantitative then you can perform Multivariate analysis. If the outcome variable is categorical and also predictor you can do odds ratio, even multiple logistic regression.