1
$\begingroup$

I am a trainee clinical psychologist investigating whether age, ethnicity, or gender influence the uptake of Cognitive Behavioral Therapy (CBT) for psychosis in four different complex needs teams. This is a small quality improvement project of 5,000 words. I have already run descriptive and chi-squared analyses on a different aspect of the data for which I received help on this forum. I have created a number of categories of uptake of therapy (DV) e.g. CBT>16 sessions, CBT ongoing, no CBT offered, etc. At a minimum I need to include 3 categories but 5 categories would be better (I can use just 3 if the analysis is going to become a lot more complicated for each additional category). Ethnicity is divided into 5 categories, I am not sure how to divide age, and gender speaks for itself.

My plan was to do a regression for each of the four teams. For example, in Team 1 do any of the ethnic categories predict uptake of therapy? Does age predict uptake of therapy? Etc.

Two of the teams have quite small numbers <40 and because of this and missing data, some of the ethnicity categories are <5. Overall there are 315 clients. I understand the power of the regression will be greater, the bigger the sample I have, but I have to balance this out with it being of greater clinical use knowing results for each team rather than the service as a whole. But I am open to amalgamating all four teams into one IV if this will make my life easier.

I have SPSS and Excel. I have carried out multiple regression before with SPSS but I just want the simplest, most straightforward way of carrying out the analysis. My hypothesis is the null one - ethnicity, age or gender do not predict the uptake of therapy.

Improve this question
$\endgroup$
7
  • $\begingroup$ Can you briefly describe the difference between the teams? What do you mean by "a small quality improvement project of 5,000 words?" $\endgroup$
    – Marquis de Carabas
    Commented Jul 21, 2015 at 15:42
  • $\begingroup$ Why do you think age, sex, or ethnicity will not have an effect on uptake? What would be the purpose of this modeling exercise if you don't think there is an effect on the outcome? Should you not be fitting the model with the variables that are of theoretical importance to your outcome? $\endgroup$
    – Marquis de Carabas
    Commented Jul 21, 2015 at 15:48
  • $\begingroup$ In terms of age sex and ethnicity, this is an exploratory analysis. There is some evidence that age, gender and ethnicity may influence uptake of therapy but it is not particularly strong e.g. in terms of ethnicity the evidence base for psychological therapy not relevant to non-western cultures. This is not the main thrust of the project which but my supervisor thought it a good idea to add on as the data was available. $\endgroup$
    – Matthew Rosen-Marsh
    Commented Jul 22, 2015 at 19:43
  • $\begingroup$ A Quality Improvement project is similar to a clinical audit and is undertaken by Trainees in one of their 6 month placements. The aim is to provide information which will help improve service delivery. $\endgroup$
    – Matthew Rosen-Marsh
    Commented Jul 22, 2015 at 19:50
  • $\begingroup$ What other variables do you have in your dataset that you suspect is associated with uptake, then? Is "Team" your main covariate of interest? $\endgroup$
    – Marquis de Carabas
    Commented Jul 22, 2015 at 19:51

1 Answer 1

Reset to default
0
$\begingroup$

The answer to your first question is quite obvious: If your outcome is categorical with >2 categories, you cannot use OLS. You need to fit a multinomial logit model.

Second, you can certainly do a stratified analysis for each team, i.e. fit:

$Uptake = \beta_0 + \beta_1Age +\beta_2Sex + \beta_{3}Ethn2+ \beta_{4}Ethn3+ \beta_{5}Ethn4+ \beta_{6}Ethn5 + \epsilon$

or you can combine the four teams into one dataset and interact team with your three model covariates:

$Uptake = \beta_0 + \beta_1Age +\beta_2Sex + \beta_{3}Ethn2+ \beta_{4}Ethn3+ \beta_{5}Ethn4+ \beta_{6}Ethn5 + \beta_7Age*Team1 + \beta_8Age*Team2+\beta_9Age*Team3+\beta_{10}Sex*Team1+...+\beta_kEthn5*Team3 + \epsilon$

And do a Chow test for "structural differences" across the 4 teams. Either way, the choice is yours. I entered age as continuous here, but you can certainly divide it into different age categories to enter into the model if it makes sense in your context.

Improve this answer
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.