Regressions and significations - p < 5%

[Pita Fouta]

Hello, I wanted to have an opinion, concerning a regression.

When p<5%, do we say that it is the :

1) the p-value that is significant ?
2) the coefficient that is significant ?
3) or the relationship between the independent and dependent variable that is significant ?

Thanks in advance;

[Nick Cox]

1) is wrong or at least usage better avoided in my view. Loosely, a P-value indicates or shows significance or its lack but it is not in itself significant. That's perhaps a fine distinction. 2) and 3) are closer to what people get away with.

In each case there are hundreds (thousands?) of textbooks and papers full of warnings against (a) taking conventional thresholds such as 5% too literally (b) taking significance tests too seriously.

Also, people have been suggesting better alternative wordings for dependent and independent variables for at least 60 years.

I have been using response rather than dependent variable for most of that time, although in my hearing and reading outcome often works well.

A better term for independent variable is more a tribal matter. I tend to talk about predictors but that's also used for the entire fitted function yielding predictions. Over time covariate has morphed from having a specific meaning in analysis of covariance to being a general term for anything not a response or outcome. Consider also explanatory variable and yet more.

If you want opinions about usage in your field, you need to state your field and/or ask people in that field.

(There is a bizarre idea floating around that everyone here is an economist, but it's not true.)

[Carlo Lazzaro]

Pita:
as an aside to Nick's detailed explanation about your query, you may want to take a look at https://www.bmj.com/content/bmj/311/7003/485.full.pdf

[Nick Cox]

As a detail: of the top 20 people on Statalist ranked by #posts since 31 March 2014 perhaps 6 or 7 are economists, starting with Carlo Lazzaro at position 3.

Of course, the correlation between chattiness and competence is probably modest.

[Carlo Lazzaro]

Nick:
I confirm both terms are modest at this end!
I spent much time when younger preferring Stan Smith (https://it.wikipedia.org/wiki/Stan_Smith) to Adam Smith (

https://en.wikipedia.org/wiki/Adam_Smith) .

[Jared Greathouse]

This should be a required piece of reading for all intro to stats courses on this subject.