Hello I was wondering whether you could help me with my analysis. I'm not sure if I've done this correctly and am basically pleading for some help with this! I hope someone could get back to me asap! as I have to submit my proposed analysis section (which is already late). I feel like bursting into tears and would really really appreciate some assistance!

I've been told to use some form of regression. I originally decided on using non-parametric tests, but was told that non-param is just not robust enough and not suitable for data analysis.

I'm looking at whether services are meeting the needs of service users according to a persons specific needs. In a sense, predicting whether services are meeting the needs of a service user.

So far, my variables are as follows:

IV: Need (individual) Did the SU have a specfic need (YES/NO)
DV: Need (service) Did the service meet this specfic need (YES/NO).

As the DV is categorical, I've decided to conduct a logistic regression equation. I've never used logistic regression and I'm not sure whether I'm going down the correct path here.

Please help!!!

You've just got a 2 x 2 contingency table right?

If so, I don't see why you can't use a Chi-Squared or Fisher's exact test?

Sorry, just read your problem more clearly and have a question ...

if the SU has no specific need, how do you define the response?

I originally did specify to use Chi-square and Fishers (if lower than 5 expectancies) but was told not to use it as the technique is not robust enough.

Basically, with my IV some groups have particular needs; thus if they had a need (with a particular question) they would have a score or value of 1. If they did not have a need-no need the participant would have a score of 0.

If a participant had a score of 1, I would be looking to see whether the service responds to this need-thus YES response from the service.

Looking to predict that if a participant has a need (score of 1-YES response) I will be looking to see whether I could predict the service to respond to that need ie the service would also have a score of 1-YES response).

Are you using SPSS? Binary logistic regression is fairly straight forward in SPSS. Analyse-Regression-Binary logistic. Remember to specify that your DP is categorical (as prompted) and enter the categories. I use the Hosmer-Lemeshow goodness-of-fit option - to test whether the model is a good fit to the data (this is basically a chi-square test, ironically - you want it to be non-significant). The 'omnibus tests of model coefficients' will give you the significance level for your predictor (again, this is given as a chi-square result). Your 'model summary' will give you a measure of the amount of variance explained by the model - I use the Naglekerke R square figure given.

An SPSS manual will give you the table output and interpretation.

Hi Pineapple, if you're using SPSS, you can download SPSS For Dummies here: http://www.mediafire.com/?x5w1nxyzdrn
It's a good introduction if you're not too familiar with SPSS.

So you are interested in the following probability ...

Pr( Service meets need GIVEN Service User has a need )

That's fantastic! Thanks very much for all your help and advice :) I'm going to be burning the midnight oil writing all of these priniciples behind LR :( Oh dear. Haven't even started on the qualitative analysis yet, but I'm ok with qualitative data :)

Yes, Jewel, that's what I'm looking at :)

it doesn't sound like you actually need logistic regression.

So you are interested in the following probability ...
Pr( Service meets need GIVEN Service User has a need )


It sounds like you need to calculate odds ratios.

I don't understand any of this! What subject is it? Here's a test for you: explain it in 20 words to a layperson outside of your field.

I maybe misinterpreting your requirements but a maximum likelihood estimate of the Pr( Service meets need GIVEN Service User has a need ) is simply the empirical ratio of the number of people with a need who are satisfied by the service to the number of people with a need.

Isn't odds ratio an important part with logistic regression? when it comes to interpreting data?

You can do this with a pen and paper, believe it or not.
-Draw a 2x2 table.
-along the top two columns, label these need- yes; need- no (need being your "exposure"
-down the side label two rows, response made - yes; response made - no (where needs met is your outcome)
-Fill out the boxes with your appropriate observations
-calculate the odds of a response when there is a need, and divide by the odds of a response when there is no need
These easiest way is to label your cells a, b, c, d: where top left is a, top right is b, bottom left is c and bottom right is d.

and perform (a/c)/(b/d)
This number gives you the odds of a response in the needy group, compared to the odds of a response in a non-needy group.

OMG - thank you so much! I've been looking for SPSS for Dummies in all the bookshops I've been to in Germany and Holland for years! Am very pleased