3rd August 2010 at 11:11 am #4244
an analysis of how governing parties/candidates (apolitical/left/center/right-wing, 4 dummy variables) effect changes in unemployment rates. I’m running a
cross-sectional, multi-variate regression analysis in SPSS where I control for
alternative explanations of unemployment such as: government effectiveness,
inflation, gdp, number of veto players, internal conflicts ect. All of the
variables have data for 100+ units (countries) and are either metric or
dicotome. Now, I know that it is not necessarily a problem that the results of
the analysis do not have significant sig.-values, but I can’t shake the feeling
that I’ve done something wrong when none of my 12 explanatory variables turn out
to explain anything. Any ideas? I’m a bit rusty when it comes to quantitative
methods so it might be some beginner’s mistake…
The dataset is attatched
Roger3rd August 2010 at 11:20 am #4250
Ps. I have checked for high levels of bi-variate correlations between the explanatory variables and none of the variables I’m using have correlations above 0.7. I can’t remember how to plot in the tolerance levels, but I suppose that it is not a major issue when the bivariate correlations are relatively low3rd August 2010 at 6:49 pm #4249Jesus TangumaMember
what kind of var is the dep var?
have you checked the biv corr between and among predic var and dep vars?
be sure to run the appropiate corr if you have metric/nonmetric vars
depending which spss version you are using, you can ask for collinearity diagnostics, residuals,,,
hope this helps,
Jesus8th August 2010 at 5:24 pm #4248Marc NauraMember
Hi, what is your R-square?
In regression analysis, there are assumptions around the normal distribution of residuals and linearity between the predicted and the explanatory variables. Have you checked for that? It could be that you have non linear relationships in your data. You can simply test it visually by plotting unemployment vs each individual variable.
Also, you need to be careful about the relative number of explanatory variable sagainst your total sample size. The general recommendation is that you should have at least 10 sampling points per explanatory variablesto avoid over fitting the data.therefore, if you are using 12 variables, you should have a sample size of 120 or more.9th August 2010 at 12:55 am #4247macrylindaMember
Welcome to this site and I am also newbie here.
My name is Disoza and I have just recently join this site.
I am here to interact with people and definitely get the information regarding experiencefestival community.
Glad to join this site.
[url=http://moviesonlinefree.biz/]watch free movies online[/url]9th August 2010 at 5:01 pm #4246
Thanks alot, I will check it out and perhaps remove some variables from the model.10th August 2010 at 9:12 am #4245
I’ve figured it out! I missed some values for employment rates when I made the dataset and eventhough all of my variables had 100+ units the collective ammount of valid units for the model was only 78. When I fixed the dataset it all worked out and the results have shown that it did not matter who governed the country between 2007-2008, at least not for employment rates.
- The forum ‘Default Forum’ is closed to new topics and replies.