need help regarding statistical methods for a project :(

[yznahc]

hi all, i have been tasked to do a biostats project and really need some advice from all the great biostatisticians out there!

i have two different treatment outcomes (Yes or No), and a whole bunch of about 40 different risk factors(some risk factors are nominal, and some are interval/ratio).

Would it be logically sound if i first did a bivariate analysis on all my respective factors using the chi square method to see if each risk factor is associated with the treatment outcome?

Following chi square test, with those factors that come out to have p<0.05 (meaning they are statistically significant), can i put these factors through multivariate logistic regression to find out their independent associations to the treatment outcome?

Any better suggestions to do the stats analysis?

Thanks for your help in advance! Greatly appreciated!

[simnasameer]

Thats absolutely right!! I did the same and my supervisor has approved it... way to go!!

[simnasameer]

1. Summary statistics/ Descriptives- frequency listing , histograms, For continuous variables (interval) means and standard deviations (95% confidence intervals) should be calculated for data that are symmetrically distributed and medians, quartiles and ranges for skewed data. etc
2. Bivariate relationship- use chi square for categorical (nominal variables). For interval variable independent t-tests or equivalent non-parametric tests need to be done (cant do chi square)
3. Whatever significant at step 2 use for multivariable analysis

[yznahc]

ok thanks for your reaffirmation simnasameer! do u happen to have any good guides to recommend on multivariate logistic regression? totally have no idea how to use that statistical method

i also read about binary logistic regression. would using binary logistic regression be any more superior/powerful and more accurate than those test u earlier mentioned?

for interval/ratio risk factors, could i also possible just stratify the factors into two groups and do a chi square? if i do so, does it mean, some 'essence' of the data is lost?

many thanks

[simnasameer]

You have to go through it in steps. So you have to do step 1 and 2 i mentioned above and that will take some time- focus on that now

Binary logistic regression is a type of multivariate analysis (step 3). That will be the preferred method for you ("Binary" because your outcome variable has "bi" or "2" measures only (yes/no).

I read David V deus - Surveys in social research- very nice book and easy to understand stats for dummies (helped me a lot)

[yznahc]

ok thanks for the recommendation on the text. sorry but now i am real confused, does it mean, i can actually skip doing chi square/unpaired t tests, and i just do binary logistic regression first, followed by multivariate logistic regression?

i am totally a stats dummy D:

[simnasameer]

No no no!!! Do step 1 and 2!!!
Binary logistic regression is step 3 (also called multivariate logistic regression)!! And you choose only those variable in step 3 that are significant in step 2 (after chi squared/t test)