Interpretation of my model

[Vincent Wey]

hello Stata-Community!

I have two questions regarding the interpretation of my OLS regression model (panel data) and hope to find an answer here!

I generally want to measure the impact of government ideology on the income-support of several OECD-countries during Covid. Income-support as the dependent variable is measured with 0(no income support), 1(mediocre income support) and 2(high income support). You can see the results in the appendix!

1. question: the variable income support has little fluctuation (at some states, income-support doesn't change at all and stays e.g. for two years at a mediocre level) - can I even make assumptions about a significant relationship between the variables if my independent variable (Gove_Reli - Government Ideology) is significant?

2. question: I added the variables GINI-Index (GINI_2019) and Poverty-Rate (PVT_2019) as predictors to the (second) model. Whereas in the first model, my independent variable (Gov_Rile) was significant, now after adding it has a p-value of .714. How can I interpret this sudden change?

Thanks very much in advance for you answers!

FYI: my other predictors are: human development index , hospital beds per thousand, gdp per capita, icu patients per million, reproduction rate, deathrate per million, incidence per million

[Carlo Lazzaro]

Vincent:
1) judging by Adj R_sq, you should go with the model whose outcome table you uploaded as your first screenshot (please note that, as per FAQ, screenshots are not recommended or openlyu deprecated on this forum. Use CODE delimiters instead. Thanks);
2) with such a large sample, I would take a look at non-default standard errors (at least for heteroskedasticity);
3) as in each and every regression, you can conclude that a chanege of one unit in a given coefficient produces an X variation in your regressand when adjusted for the other independent variables;
4) I would run -linktest- to check for the correct specification of your regression model.

[Vincent Wey]

Hello Carlo,

thanks very much for your really helpful answer. I checked for heteroeskedasticity (I did the Breusch-Pagan-Test for each variable) and came to the conclusion that every one of my variables have heteroeskedatic distribution. So I took robust standard errors for my model.

Besides, since I have panel data, I assume I have to use the according longitudal linear regression (xtreg) for my data?

Another issue that came to my mind was that I have a ordinal scaled dependend variable (0,1,2). But since it is quasi-symmetric (0 for no income support; 1 for medium income support; 2 for high income support), i guess I can proceed with the linear regression? Or do I have to think about a logistic regression model?

Thanks again for the answer in advance!

Best regards

Vincent

[Carlo Lazzaro]

Vincent:
if you have panel data with an ordinal dependent variable, you should consider -ologit- with clustered standard error.

[Vincent Wey]

Even if my ordinal dependent variable can be considered, as mentioned above, quasi-metric?

Best regards

Vincent

[Carlo Lazzaro]

Vincent:
most depends on the tribal rules in your research field.