Anyone studied regression?

Barks · 19 May 2009 at 12:19

Hi,

Please note I'm not asking for anyone to actually do the questions below, I'd just really appreciate some help in which direction I should take to answer them. I've been revising for the last few days and understand most of the concepts, but I'm completely stuck on this:

I can deduce that R = 76% so overall the variability in the y variable is explained by the independent variables.
But how do I test the significance of each of the independent variables. They haven't provided enough data. I need things like SSR, SST, SSE, MSR, S etc.

I'd really really appreciate if someone could point me in the right direction here, perhaps part b) is actually very straight forward and I'm completely missing the point.

Many thanks,

Anominity · 19 May 2009 at 12:29

I have studied Regression, first I need to introduce the hypnotic trance on you, then get you to regress back to when you were a child or another time, usefull for working out ISE's and SSE's with regards to Phobia's or depression and the like.

Oh wait, something tells me thats not the regression you are talking about :S

muon · 19 May 2009 at 12:30

Use the t-statistics (the ones in parentheses) and t-distribution tables to get a significance level for each coefficient.

Table: http://www.socr.ucla.edu/Applets.dir/T-table.html

I wish my econometrics exams were this easy. (3rd year econ student here

)

Barks · 19 May 2009 at 12:31

ghost101 said:
Use the t-statistics (the ones in brackets) and t-distribution tables to get a signifaicance level for each coefficient.

I feel like an idiot now, had absolutely no idea they were t statistics.

Many thanks, i'm sure you can see now why I thought it was impossible to answer

I presume for C I used an F test instead.

Cheers

Barks · 19 May 2009 at 12:34

Final question, how do I determine which value of a (significance level) to use?

Cheers

muon · 19 May 2009 at 12:35

scott212 said:
Final question, how do I determine which value of a (significance level) to use?

Cheers

Thats a subjective choice. But usually at least 5%. If it is more significant, mention it.

And yes F-test.

Barks · 19 May 2009 at 12:38

ghost101 said:
Thats a subjective choice. But usually at least 5%. If it is more significant, mention it.

And yes F-test.

So the examiner would not mark me down if I picked a=0.01 or a=0.05?

Many thanks for your help mate

muon · 19 May 2009 at 12:40

scott212 said:
So the examiner would not mark me down if I picked a=0.01 or a=0.05?

Many thanks for your help mate

5% is adequate to suggest significance for my courses. If you use 1% across and therefore conclude that the coefficient is insignificant when it is to the 5% level, then you are likely to be marked down a bit. Depends on the person who is marking I guess if they do and your lecturer's policy.

Barks · 19 May 2009 at 12:47

The test statistic follows a t distribution with (n-2) degrees of freedom. In this case, as I'm calculating the significance for each independent variable, what value would I use for the df?

This stuff is extremely confusing when you haven't grasped it

monkentage · 19 May 2009 at 12:53

sample size - 2

Barks · 19 May 2009 at 12:56

monkentage said:
sample size - 2

where does this come from though, at the top it says n=36.

does this mean I use a df value of 8?

monkentage · 19 May 2009 at 13:18

you said n-2 degrees of freedom. if you wanted to know what df for the regression as a whole i think it would be n - k - 1 where k = number of parameters

Barks · 19 May 2009 at 13:28

monkentage said:
you said n-2 degrees of freedom. if you wanted to know what df for the regression as a whole i think it would be n - k - 1 where k = number of parameters

I'm confused. In the textbook it just says "test statistic follows a t distribution with (n-2) degrees of freedom" which I assumed was a general rule?

So what value of DF do I use when conducting a t-test for each individual independent variable, as in question b)

Cheers

muon · 19 May 2009 at 14:21

scott212 said:
The test statistic follows a t distribution with (n-2) degrees of freedom. In this case, as I'm calculating the significance for each independent variable, what value would I use for the df?

This stuff is extremely confusing when you haven't grasped it

It will be 36 - number of parameters calculated. Your textbook is probably using the basic example.

Revision presentations

http://econ.lse.ac.uk/courses/ec220/G/ieppt/series2/

Covers loads of topics. Its for the basic metrics course at LSE.

Barks · 19 May 2009 at 14:24

I see, so the -2 is not a general rule but the number of parameters. Should be (sample size-number of parameters)

So finally, as I'm only calculating the significance of each independent variable, the number of parameters would be 2 for each t-test I do?

Cheers

Barks · 19 May 2009 at 15:17

delete

Barks · 19 May 2009 at 16:33

Absolute last question, promise.

1) Is the overall regression significant? Conduct a formal test and interpret your results.

2) How would you test for the hypothesis that joint impact of unemployment rate and competitiveness on the UK balance of payments is insignificant?

What's the difference in answering these. I'm aware Q2 is an F-Test. For Q1 I've already worked out that some of the T values are significant and some aren't, so by answering the question do I just add these together or something? Or do both questions require F tests?

Thank you very much for your help, those lecture slides are fantastic... have really helped me crack on with the revision.

Amp34 · 19 May 2009 at 16:47

Yeah I have, it normally results in a higher erosion rate and lower sea levels...

Oh wait, not that kind of regression...

muon · 19 May 2009 at 17:52

scott212 said:
Absolute last question, promise.

What's the difference in answering these. I'm aware Q2 is an F-Test. For Q1 I've already worked out that some of the T values are significant and some aren't, so by answering the question do I just add these together or something? Or do both questions require F tests?

Thank you very much for your help, those lecture slides are fantastic... have really helped me crack on with the revision.

There are 2 methods for the F Test. One which measures the explanatory power of the whole model, and then one the explanatory power of a group of variables.

http://econ.lse.ac.uk/courses/ec220/G/ieppt/series2/C3D7.pps#383,25,Slide 25

Slide 25 onwards describes the latter. 1-25 describe the former.

Make sure you do well in your metrics module if you want to get into a good masters course. It is possibly the most important area to be competent in if you wish to pursue economics at a higher level.

Barks · 19 May 2009 at 19:21

ghost101 said:
There are 2 methods for the F Test. One which measures the explanatory power of the whole model, and then one the explanatory power of a group of variables.

http://econ.lse.ac.uk/courses/ec220/G/ieppt/series2/C3D7.pps#383,25,Slide 25

Slide 25 onwards describes the latter. 1-25 describe the former.

Make sure you do well in your metrics module if you want to get into a good masters course. It is possibly the most important area to be competent in if you wish to pursue economics at a higher level.

Thank you very much for your help and advice.