Which analysis method should I choose ? Need your help

[yasmine yasmine]

Hello ,

I have :

- dependent variable (Y) which is continuous
- four independent variables (X) which are dichotomous (0/1)
- four continuous control variables which are mixture.

What method should I choose to test the influence of X on Y? multiple or logistic regression? and which tests must i do ?
NB. I have a balanced panel

thank you in advance

[yasmine yasmine]

and if i must use one-way ANOVA, so there are any specific tests to do? like heteroscedasticity etc. !!
and how can i interpret the results ?
is there a reference to follow it please ?

[Carlo Lazzaro]

Yasmine (as per FAQ, please note the preference for family names, too on this forum):
I would take a look at -xt- and -xtreg- entries in Stata .pdf manual.
Reading a textbook on panel data analysis is recommended as well (please, see http://www.stata.com/bookstore/stata-press-books/).

[Marcos Almeida]

Hello Yasmine,

Carlo gave excellent suggestions. Precisely, and now quoting you, for the vast majority of Stata users (if not all), I believe the manual is the "reference to follow".

You may type - help regress -, - help anova - and help xtreg - in the command window of your Stata, and get the link to the manual.

Best,

Marcos

[yasmine yasmine]

Hello,

Thanks a lot for both of you Mr Carlo and Mr Marcos;

But i think that before choosing the manual i must understand what method should i choose !

Can someone help me please ?

as i am far of statistics, i do not know if my dependent variable follows a binomial or gaussian distribution.
I attached here my variable Y and the Gauss curve obtained on Excel
My dependent variable is not a percentage. In fact, it is the sum of 3 residues coming from three different regressions. It (Y) contains positive and negative values and ranges from -53.84 (Min) and 15.76 (Max) and that my balanced panel covers 180 firms and 14 years.

Thanks,

[Carlo Lazzaro]

yasmine (as per FAQ, please note the preference for family names, too on this forum):
as Marcos already recommended you, especially if you are "far from ststistics", the first step to take is taking a look at the suggested Stata manual entries.
This will ease your task concerning what method fit better with your data.
That said, in your case -xtreg- would seem the way to go; however, simply copy and paste a string of Stata code won't get you very far: that's way a theoretical background is mandatory for dealing with this kind of stuff effectively.

[Marcos Almeida]

I couldn't agree more with Carlo's comments. There is no safe way to bypass the first steps when one is "far of statistics". And reading the manual is part and parcel of the learning process to everyone in the field. There, Yasmine will find not only rather practical examples, but also the theoretical background to start up the reflection on choosing models. That said, as Carlo also underlined, it seems Yasmine's most suitable model might well turn out into a panel data. Being so, I fear such model wouldn't be the easiest way of getting to grips with stats. First things come first: for example, getting to understand patterns of distribution of variables as well as how to cope with them, and the assumptions of the "basic" tests, such as Student's t, ANOVA and linear regression. Then, panel data under - xtreg - would become an "extension" - xt - of the regression analysis and its intricacies less mind-boggling. Indeed, so many concepts are quite difficult to be abridged, say, "in a nutshell". To end, the "composite" dependent variable (sum of 3 different regression residuals to be applied in a panel data regression) may be something people would be reticent about.

[yasmine yasmine]

I couldn't agree more with Carlo's comments. There is no safe way to bypass the first steps when one is "far of statistics". And reading the manual is part and parcel of the learning process to everyone in the field. There, Yasmine will find not only rather practical examples, but also the theoretical background to start up the reflection on choosing models. That said, as Carlo also underlined, it seems Yasmine's most suitable model might well turn out into a panel data. Being so, I fear such model wouldn't be the easiest way of getting to grips with stats. First things come first: for example, getting to understand patterns of distribution of variables as well as how to cope with them, and the assumptions of the "basic" tests, such as Student's t, ANOVA and linear regression. Then, panel data under - xtreg - would become an "extension" - xt - of the regression analysis and its intricacies less mind-boggling. Indeed, so many concepts are quite difficult to be abridged, say, "in a nutshell". To end, the "composite" dependent variable (sum of 3 different regression residuals to be applied in a panel data regression) may be something people would be reticent about.

Thanks a lot for you;

I have knowledge about multiple regressions
Now I made changes to my model:
I ended by choosing 5 explanatory variables on which 3 are binary and 2 are continuous
I posted another topic on the forum and I hope that I get help because I have a time constraint

Thanks for your help