Re: interpretting F & t-vals from regression

Rob Campbell wrote:

<snip>
I'm not sure I entirely follow you. I think it's because I should have
sent more information. Here it is.

You say you are testing for differences from zero. However, it is the
I was saying that with reference to the t-tests but you're right below:
these are treatment contrasts so they may not be differences from zero.
But in this case I think they are (see below).

I have subjects (factor) each doing 5 durations (a factor). So these
aren't crossed: durations are nested within subjects. Then within
durations each subject does 5 intensities, which I treat as a continuous
variable because I'm interested in the slope.

I fit a mixed-effects linear model with lme. I have a random intercept for
durations within subjects (random=~1|subject/duration)

anova gives me:
(Intercept) 1 165 115.70888 <.0001
duration 4 12 30.60556 <.0001
duration:intensity 5 165 9.65925 <.0001

The anova() function in R is reporting sequential tests.
That is the test for duration:intensity is after other
terms are in the model.

One thing that caught my eye is that (unless you trimmed it)
you have fit the interaction between duration and intensity without
including intensity.

and summary shows:
Value Std.Error DF t-value p-value
duration40:intensity 0.078111 0.0533651 165 1.463718 0.1452
duration100:intensity 0.086904 0.0533651 165 1.628480 0.1053
duration200:intensity 0.029069 0.0533651 165 0.544725 0.5867
duration500:intensity 0.008720 0.0533651 165 0.163403 0.8704
duration1000:intensity 0.006706 0.0533651 165 0.125671 0.9001

These are "tests" for different levels of a categorical variable.
I don't know how to interpret the coefficients in the absence
of an intensity main effect.

pairwise comparisons among the levels of your factors (k) that is tested
by the F test. The number of comparisons is actually: k(k-1)/2. If I
have a factor with 4 levels, then I have 6 possible comparisons.

The coefficients from summary.lme are the slopes and they are not with
reference to anything so I guessed that the t-tests must be for a
significant difference from zero slope. As you can see, none are
significantly different from zero.

So this high F-value is confusing me. Are you saying that it indicates a
significant difference in the estimated slopes of intensity across
duration? Even though the estimates themselves are so uncertain they
cannot be considered to be significantly different from zero? That
seems... odd.


So I'm safe in reporting the result as "none of the 5 durations tested had
intensity slopes that were significantly different from zero" ?
I ignore the F-value because it's not the hypothesis of interest?

Rob

--
Kevin E. Thorpe
Assistant Professor, Department of Public Health Sciences
Faculty of Medicine, University of Toronto

.

Relevant Pages

  • Re: interpretting F & t-vals from regression
    ... these are treatment contrasts so they may not be differences from zero. ... I have subjects each doing 5 durations. ... significant difference in the estimated slopes of intensity across ...
    (sci.stat.consult)
  • Re: Tracking Time Spent
    ... >The only way I can think of it to keep track of minutes and when it reaches 60 to add the hour and zero out the minutes. ... date/time field in the textbox named txtTaskTime. ... Note that Date/Time values are not really appropriate for durations; ... Come for live chats every Tuesday and Thursday ...
    (microsoft.public.access.formscoding)