p-value of a coefficient from a regression

oahz

New Member
#1
The t-stat value follows a two-tail symmetric t-distribution with mean and median of zero?

Am I correct so far?

If so, then how can you get a p-value > .5 if the t-stat is greater than 0?

What does a p-value of .89 mean? Wouldn't that mean the t-stat is way negative and is to the other end of the tail?
 

hlsmith

Omega Contributor
#2
Not following, may help if you provide your example. The p-value is the probability of getting a value (parameter) that extreme given that there really exist not relationship. Are you talking about your t-statics or p-values? T-statistic is the difference over the standard error, which is the number of standard deviations the groups differs from the other group. So typically if it is around 2 standard deviation different in either direction, we say the probability is low for that extreme of a difference under the pretenses that the two groups are equal.
 
#4
Oahz, I think you are confusing the relationship between the t-statistic and p-value.
I think visuals can help understand distributions and p-values, so to build on hlsmith answer, here are a few links:
Even though this article is based on the premises that you are using minitab (a stats program) - the basic theory is there: http://blog.minitab.com/blog/statis.../what-are-t-values-and-p-values-in-statistics.

Here is a video series that may help also: https://www.khanacademy.org/math/pr...sis-testing/v/hypothesis-testing-and-p-values
 

oahz

New Member
#5
I think might have figured it out. Tell me if I'm correct.

My question was, how is the p-value for an estimated coefficient derived?

Answer:

If you get a positive t-stat of, say t, and t corresponds to a probability, say p*, of randomly getting t or a value greater than t from the t-distribution, then the p-value is equal to 2p*.

Conversely,

If you get a negative t-stat of, say t, and t corresponds to a probability, say p*, of randomly getting t or a value less than t from the t-distribution, then the p-value is equal to 2P*.

Am I correct?
 

hlsmith

Omega Contributor
#6
Are you examining a simple linear regression model? If you tell us about the model we can provide more and better information.
 

hlsmith

Omega Contributor
#8
You are testing whether the variable has a slope equal to zero.


It is taking the slope value (coefficient) for the variable and dividing it by the standard error (sampling variation) and that provides the t-statics. Can be negative or positive, just like the slope. Now it interprets the value based on the degrees of freedom using the t-distribution. If the t-statistic is large or small enough then it would have a low probability of occurring given that the slope is really zero (null hypothesis).
 

oahz

New Member
#9
You are testing whether the variable has a slope equal to zero.


It is taking the slope value (coefficient) for the variable and dividing it by the standard error (sampling variation) and that provides the t-statics. Can be negative or positive, just like the slope. Now it interprets the value based on the degrees of freedom using the t-distribution. If the t-statistic is large or small enough then it would have a low probability of occurring given that the slope is really zero (null hypothesis).
I understand that much.

My confusion is the p-value.

Just answer this one question: can you have p-value of greater than .5 if the t-statistic is greater than 0?