I know the question is a little bit confusing, so let me explain it. I am writing a paper to discover factors that affect poverty alleviation. It's panel data and collected at household level.
Dependence variable: whether a household poor or not poor
Independence variable:
Remittances, Crop choices, Self-employment, Wages from manufacturing, and Government support programs
I have some trouble with the model as I don't know how to put the "Government support programs" variable into the models.
The government has some support programs such as send people aboard to work (so they can have remittance), send experts to farmers to consult (better crop choice), microcredit (so they can be self-employed), and training so they can get into the manufacture sectors. In the questionnaire, they ask people whether you receive these kinds of support so I have all the data about it.
One of my goals is to know how much impact of each factor (remittance, crop choice, self-employment, wages from manufacturing) on poverty comes from the support of the government program. So I can access the efficiency of these programs. But I can't find the way to put it in the models as it will likely correlated with other independent variables. Do you have solution to this problem?
Thank you very very much in advance!
The explanatory variables are likely to have some correlation. Check the VIFs to ensure that the correlation is not high to cause multicollinearity in the model. The problem will arise if the program is constant for a household for all years (the it will be dropped from panel data model). Try adding random intercepts or interaction with other independent variables.
can refer to Evaluating Effects of Independent Variables by
Michael S. Lewis-Beck and Lawrence B. Mohr
Hi Bach Vu,
First, your trouble with the model you could solve if for instance you create dummy variables. Since you have the answers from the people whether they receive or not government support, simply you create dummy variable that has a value of either 0 or 1 for any observation in your sample, i.e, having 1 for observations that have received government support and 0 if not. Another solution may be to create grouping variables if you have more than two outcomes and your variable is a discrete variable. Second, regarding the main question you have, the colleagues have provided some relevant answers or options how to do it. Simplistic way is also to form a group of all of these variables (independent and dependent) and then to look for correlation item and you will receive the results as you wish. Or, you could try with using factor model/analysis and you will get all the correlation you needed and much more of that. Since you want to explore the impact of each factor on poverty this would be best option for you to do because it provides all you want to get. Good luck and best wishes with your research!
Félix Sixto Pilay Toala
Economista graduado en la Universidad Central del Ecuador. Magister en Administración Pública. Candidato a Doctor en la Universidad Internacional Iberoamericana de México. Ha sido Director, docente a tiempo completo de la Pontificia Universidad católica del Ecuador sede Manabí. Docente a medio tiempo de la Universidad Estatal del Sur de Manabí y Director y Profesor a tiempo completo de la Escuela de Ciencias Administrativas y Contables de la Pontificia Universidad Católica del Ecuador Sede Santo Domingo, ha sido Profesor por 20 años consecutivos, en las asignaturas de Economía, Macroeconomía, microeconomía, Investigación, responsabilidad Social. Sus libros publicados: Salario Inflación y Conflictos Sociales y Presupuesto participativo, para una efectiva participación ciudadana, son comercializados por Amazon y varios comercios electrónicos en el mundo. Ha escrito varios artículos científicos publicados por revistas indexadas al Latindex y abarcan temas económicos y administrativos. Ha participado como profesor invitado, ponente y conferenciante en temas de responsabilidad social y liderazgo en varios Congresos a nivel de Latinoamérica así como en Europa.
You may have to use partial dependency instead of total dependencies. For example, where we were using total derivatives, we have to use partial derivatives with respect to that variable.
What kind of data do you have (e.g. time series, panel data, etc.)?
I suggest simultaneous or SUR regression models if you have bidirectional causality or endogeneity problem. I'm not sure I totally undertood your model but I think your problem is not multicoliarity but endogeneity. Besides your dependant variable is binary (poor or not poor). So the VIF option to detect correlation between independant variables is not possible. You have also the strutural equations possibility.