1

So is "P for trend" the same in principle as a common-or-garden P-value, but in a specific application i.e. odds ratios?

And can P-value, P for trend, and P-trend be regarded as interchangeable conceptually, terminologically, and in terms of statistical significance (at least)?

A pointer to an appropriate and citeable reference would be most useful, particularly if it states any differences or confirms sameness.

Edit responding to whuber's moderation

@whuber - thanks, but wouldn't an exact duplicate at least mention the topic of my question. I can see no mention of 'trend' or 'P for trend' on that page you provided. If on the other hand you know that P for trend is the same thing, then is there a citeable reference you can provide? Just to be clearer it is the use of the word 'trend' that confuses me, not the use of letter 'P'.

Another edit responding to whuber's moderation

Sorry if my question was not well formed. In attempting to provide my own answer in the hope of a confirmation, I think that the word 'trend' when talking about P-trend (e.g. for a reported result in the literature) sets the context as a P-value test for linear trend in odds ratio in logistic regression, against a categorical predictor variable. If that is correct, then this is described in standard texts as 'test for trend', in which case my copy of Kirkwood & Sterne provides a citeable reference. Hopefully this addresses whuber's concerns. Whilst the correct interpretation of P-value is very important, I was missing the context of logistic regression.

Big Old Dave
  • 145
  • 1
  • 6
  • I might add I have seen P for trend used where significance is 'weak' and where '...P for trend <0.0005..', so I'm not sure that helps me. – Big Old Dave Jul 07 '16 at 14:18
  • @whuber - thanks, but wouldn't an exact duplicate at least mention the topic of my question. I can see no mention of 'trend' or 'P for trend' on that page you provided. If on the other hand you know that P for trend is the same thing, then is there a citeable reference you can provide? Just to be clearer it is the use of the word 'trend' that confuses me, not the use of letter 'P'. – Big Old Dave Jul 07 '16 at 14:47
  • One moderation principle I apply is to consider what kinds of questions might additionally be allowed if one particular variation of an existing question were considered different. In this case, when I contemplate the infinity of possible questions of the form "what does 'p for odds ratio' mean," "what does 'p for a quadratic term mean'," "what does p for 'intercept' mean," I find it easy to conclude that making all these into separate questions would be pointless. The duplicate answers the entire set of questions by explaining what a p-value means for a general statistical procedure. – whuber Jul 07 '16 at 16:48
  • 1
    This Q&A helped with my second edit above http://www.talkstats.com/showthread.php/56730-Test-for-linear-trend-in-odds-ratio-in-linear-regression – Big Old Dave Jul 07 '16 at 18:43
  • 1
    ...as did this http://stats.stackexchange.com/questions/183518/calculate-p-trends – Big Old Dave Jul 07 '16 at 18:53

0 Answers0