I intend to determine the influence of some independent variables (age, household size, market distance, participation in social organizations, extension visit, livestock ownership, farm size, total income, access to credit, educational status and access to climate change information) on the choice of livelihood strategies engaged in by men and women. The dependent variables here are choice of livelihood strategies (coded as farm (Y0), farm + off-farm (Y1), farm + non-farm (Y2), farm + off-farm + non-farm (Y3)) and gender (coded as male (1), female (0)). I will like to know if any of the aforementioned independent variables determines the choice of livelihood strategies of men and women differently.
I am not sure that I fully understand your question, but the theory of Bayesian networks and associated software is provide ideas and quantitative tools that can be used to make inference on cause, effect and influence in situations where number of variables interact.
This MAY be appropriate: multivariate logit modeling as it will handle multiple predictors and multiple (Binary) responses; I have written about this before
see my response to this question on Rgate
https://www.researchgate.net/post/How_can_I_analyse_data_that_is_not_mutually_exclusive
Dear Ifeoma,
to mee it sounds as you could order you dependent variable livelihood strategies
Y0 < Y1 < Y2 < Y3
If this is the case, then you could work with an ordinal regression model. If not, then you must work with a multinomial regression model. Both models a univariate models. But from your description, I would say that you really just have a univariate problem. I would think, that gender should be a further independent variable, which may have several interactions with your other independent variables. To mee it would make sense, that gender also influences the livelihood strategie. But one may doubt that your other independent variables can predict gender!?
Best, Stephan
https://en.wikipedia.org/wiki/Ordinal_regression
Statistically speaking it is recommends to use Canonical Correlation Analysis to measure association between a set of dependent variables and two or more independent variables
But this does apply in your case because it is wrong to treat gender as dependent variable. Iin fact you have one categorical and a set of explanatory variables including gender therefore it is recommended to use multiple discriminant function MDF
I concur with Abdulrazzak Charbaji in thinking that canonical correlation would be a goot tool here. You speak of "dependent" variables. If by that you mean that some design variables influence the dependent variables in a causal way, Abdulrazzak is right: there is no way any design variable will change your gender! But I think a softer interpretation is possible, much as in regression: canonical correlation measures linear association between your two groups of variables, without any claims regarding causal direction.
I think if you interpret results in the right way canonical correlation is adequate here.
see my paper Vinod, H. D. “Canonical Ridge and Econometrics of Joint Production,” Journal of Econometrics, Vol. 4, June 1976, pp. 147-166
associated R code is given in my book snippets which are a part of the CD coming with
the book:
“Hands-On Intermediate Econometrics Using R: Templates for Extending Dozens of Practical Examples.” (2008) World Scientific Publishers: Hackensack, NJ.
(http://www.worldscibooks.com/economics/6895.html) Some free stuff is available at the
book website for exercises for the book
I agree with some of the responses above that your application sounds appropriate for a canonical correlation analysis, which is a type of general linear model. Here is an article: http://midag.cs.unc.edu/pubs/papers/AmericanStat82_Muller.pdf
It strikes me that gender belongs in your group of independent variables, but not in the same way as age, household size etc. This suggests to me two canonical correlation analyses, one for men and one for women, to see differences in dependency between the two sets of variables.
The advantages of the multivariate multilevel model are several fold - it is regression not correlation and has predictor variables and regression-like coefficients ; it can (based on rather forgiving assumptions) deal with missing values on the responses; it can deal with a mixture of different types of scale on the response including discrete and continuous; it gives the conditional correlation between the responses taking account of predictor variables; it can operate at more than one level so responses can be seen as nested within individuals, households and communities, etc; it can deal with missing by design surveys where not all people are asked the same questions..
Ifeoma - I am not sure how you are measuring livelihood strategies and whether they are measured in the same way for men and women. It would be helpful if you can explain this.