# "Decreasing P-value" could be interpreted as "Getting Meaningful?" in time series?

#### Licky

##### New Member
Hello. I'm Lee, a candidate for a master's degree in business administration in South Korea.

I'm analyzing economic data in emerging markets such as Asia, And I found that many indicators are following incumbent markets such as U.S. or E.U.

And the basis for that is the p-value in simple regression analysis between focal variables is significantly decreasing under 0.001 as the graph below.

It could be intuitively understandable, and too natural to make questions about it.

But I failed to find out articles that this kind of decreasing p-value can be interpreted as getting meaningful over time, especially in social science.

If you know anything about it such as the possibility for those interpretations or some related articles, I sincerely ask you to let me know.

With my best regards, Lee.

#### hlsmith

##### Less is more. Stay pure. Stay poor.
Decreasing p-value means that you have multiple pvalues taken from multiple tests and they seem to be decreasing? If so, better explain what you are doing.

Or

That you fit a simple linear regression and the p-value for the slope of the dependent variable regressed on time is small?

#### noetsi

##### No cake for spunky
Personally I don't assign much value to p values - which I think is the norm these days. A lower p value does not mean one measure is better than another.

#### noetsi

##### No cake for spunky
I might point out that 1) I am not the expert others are here and 2) there is disagreement among many in statistics and I only can comments on the small number of articles I have read that addressed this. Which is a tiny portion of the total I am sure. I think a change in variance over time would change your p values even if the true effect size was identical.

#### Licky

##### New Member
Decreasing p-value means that you have multiple pvalues taken from multiple tests and they seem to be decreasing? If so, better explain what you are doing.

Or

That you fit a simple linear regression and the p-value for the slope of the dependent variable regressed on time is small?
Thanks for replying! Well, yes it is multiple tests with the same samples and same variables. The only difference is the year.
There are hundreds of corporation datasets and 2 financial data(2 Variables, A and B).
For example in the graph I attached, the effect of A on B seemed nothing significant in 2012(Almost 0 Beta, 0.9 p-values) but it goes much bigger Beta and smaller p-value over time.
So in this case, if there's no problem in other areas such as theoretical background, can we say this situation "The effect A on B is turning into more statistically meaningful and significant effect over time"?
Sorry for my lack of explanation and thank you again.

#### Licky

##### New Member
I might point out that 1) I am not the expert others are here and 2) there is disagreement among many in statistics and I only can comments on the small number of articles I have read that addressed this. Which is a tiny portion of the total I am sure. I think a change in variance over time would change your p values even if the true effect size was identical.
I appreciate your opinion. What I understood about p-value is p-value shows us "the possibility of exceptional events". So I thought decreasing p-value is decreasing exceptionality which means getting more meaningful and general to the explanation for phenomenons.
Especially like the sample I attached contains a dramatic change(0.9 to 0.0)

#### noetsi

##### No cake for spunky
P values are defined differently by frequentist (the regression most study I suspect) and Bayesians according to my professors anyhow. I think p values in practice tell you how likely it is that the effect size you find will be true in the larger population although formally they are not defined that way. I think the formal definition is how likely it is you will find a value as extreme as the one you did if the null is true. Which makes a variety of assumptions including about the distribution. If the variances at different points in time are different I don't think you can compare p values at all.

Regardless its not a good idea to compare how important an effect size is by how small or large the p value is. The move is to focus more on effect sizes and less on p values.

#### noetsi

##### No cake for spunky
Incidentally time series regression creates a whole special set of problems for p values including autocorrelation, structural breaks, and non-stationarity. Good luck if you are working with that

Just say no to time series

#### Licky

##### New Member
Incidentally time series regression creates a whole special set of problems for p values including autocorrelation, structural breaks, and non-stationarity. Good luck if you are working with that

Just say no to time series
I appreciate your kind reply noetsi. It's a great help form me.
I hope you have a great end of the year!