Hi,
I am analyzing a data set with greater than 1000 independent cases
(collected in an unrestricted manner), where each case has 3 variables
associated with it: one, a factor variable with 0/1 levels (called XX),
another factor variable with 8 levels (X) and a third response variable
with two levels (Y: 0/1). I am trying to see if X1 has an effect on the
relationship between X2 and the proportion of 1-s in Y.
I have three questions:
a) I have never used glm-s for this or any other sort of analysis before
today, so am I interpreting the output correctly ?
After setting
options(contrasts=c("contr.treatment","contr.poly"))
I did:
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~Begin R output~~~~~~~~~~~~~~~~~~~~~~
Call:
glm(formula = Y ~ X * Fac, family = "binomial", data = mat, subset =
sactype < 3 & numstim == 16)
Deviance Residuals:
Min 1Q Median 3Q Max
-2.232 -0.901 0.416 0.985 1.656
Coefficients:
Estimate Std. Error z value Pr(>|z|)
(Intercept) 2.405 0.209 11.52 < 2e-16 ***
X2 -2.511 0.293 -8.57 < 2e-16 ***
X3 -3.283 0.286 -11.47 < 2e-16 ***
X4 -2.009 0.302 -6.65 3e-11 ***
X5 -3.098 0.276 -11.22 < 2e-16 ***
X6 -2.580 0.288 -8.97 < 2e-16 ***
X7 -3.484 0.288 -12.09 < 2e-16 ***
X8 -2.811 0.328 -8.56 < 2e-16 ***
Fac -1.558 0.721 -2.16 0.03071 *
X2:Fac 2.133 0.942 2.26 0.02351 *
X3:Fac 1.848 0.932 1.98 0.04748 *
X4:Fac 2.836 0.982 2.89 0.00386 **
X5:Fac 3.263 0.945 3.45 0.00056 ***
X6:Fac 3.630 0.971 3.74 0.00018 ***
X7:Fac 3.256 0.883 3.69 0.00023 ***
X8:Fac 3.350 1.000 3.35 0.00081 ***
---
Signif. codes: 0 `***' 0.001 `**' 0.01 `*' 0.05 `.' 0.1 ` '
1
(Dispersion parameter for binomial family taken to be 1)
Null deviance: 1619.4 on 1178 degrees of freedom
Residual deviance: 1271.2 on 1163 degrees of freedom
AIC: 1303
Number of Fisher Scoring iterations: 5
~~~~~~~~~~~~~~~~~~~~~~~~End R output~~~~~~~~~~~~~~~~~~~~~~~~~~~
I am reading this like this: each of the X2....X8 terms tell me whether
the proportions associated with those factors at level 0 of Fac, are
different from the proportion associated with factor X1 for level 0 of
Fac. And each of the terms associated with Fac (X2:Fac,.......X8:Fac) is
telling me whether the difference between X2...X8 and X1 is different
for Fac=0 and Fac=1; and this is the same thing as whether the
proportion associated with X2......X8 are different for the two levels
of Fac. So these X2...X8:Fac terms are like performing a simple 2x2
analysis of the effect of Fac on Y, given X2 (....X8).
How much of this is incorrect ?
My other two questions are:
b) Is this the right way to approach this analysis in R ? Or am I better
off reading about multi-way contingency table analyses and using them ?
and
c) How do I incorporate a correction for multiple-testing into the above
analysis ? The effect of Fac on the relationship between X and Y was
planned.
I would greatly, and respectfully appreciate all pointers, tips and
admonitions.
Thank you !!!!
Suresh
Oops, I corrected some errors in the first paragraph; sorry for the
repeated posting.
Suresh
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Hi,
I am analyzing a data set with greater than 1000 independent cases
(collected in an unrestricted manner), where each case has 3 variables
associated with it: one, a factor variable with 0/1 levels (called Fac),
another factor variable with 8 levels (X) and a third response variable
with two levels (Y: 0/1). I am trying to see if Fac has an effect on the
relationship between X and the proportion of 1-s in Y.
I have three questions:
a) I have never used glm-s for this or any other sort of analysis before
today, so am I interpreting the output correctly ?
After setting
options(contrasts=c("contr.treatment","contr.poly"))
I did:
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~Begin R output~~~~~~~~~~~~~~~~~~~~~~
Call:
glm(formula = Y ~ X * Fac, family = "binomial", data = mat, subset
sactype < 3 & numstim == 16)
Deviance Residuals:
Min 1Q Median 3Q Max
-2.232 -0.901 0.416 0.985 1.656
Coefficients:
Estimate Std. Error z value Pr(>|z|)
(Intercept) 2.405 0.209 11.52 < 2e-16 ***
X2 -2.511 0.293 -8.57 < 2e-16 ***
X3 -3.283 0.286 -11.47 < 2e-16 ***
X4 -2.009 0.302 -6.65 3e-11 ***
X5 -3.098 0.276 -11.22 < 2e-16 ***
X6 -2.580 0.288 -8.97 < 2e-16 ***
X7 -3.484 0.288 -12.09 < 2e-16 ***
X8 -2.811 0.328 -8.56 < 2e-16 ***
Fac -1.558 0.721 -2.16 0.03071 *
X2:Fac 2.133 0.942 2.26 0.02351 *
X3:Fac 1.848 0.932 1.98 0.04748 *
X4:Fac 2.836 0.982 2.89 0.00386 **
X5:Fac 3.263 0.945 3.45 0.00056 ***
X6:Fac 3.630 0.971 3.74 0.00018 ***
X7:Fac 3.256 0.883 3.69 0.00023 ***
X8:Fac 3.350 1.000 3.35 0.00081 ***
---
Signif. codes: 0 `***' 0.001 `**' 0.01 `*' 0.05 `.' 0.1 ` '
1
(Dispersion parameter for binomial family taken to be 1)
Null deviance: 1619.4 on 1178 degrees of freedom
Residual deviance: 1271.2 on 1163 degrees of freedom
AIC: 1303
Number of Fisher Scoring iterations: 5
~~~~~~~~~~~~~~~~~~~~~~~~End R output~~~~~~~~~~~~~~~~~~~~~~~~~~~
I am reading this like this: each of the X2....X8 terms tell me whether
the proportions associated with those factors at level 0 of Fac, are
different from the proportion associated with factor X1 for level 0 of
Fac. And each of the terms associated with Fac (X2:Fac,.......X8:Fac) is
telling me whether the difference between X2...X8 and X1 is different
for Fac=0 and Fac=1; and this is the same thing as whether the
proportion associated with X2......X8 are different for the two levels
of Fac. So these X2...X8:Fac terms are like performing a simple 2x2
analysis of the effect of Fac on Y, given X2 (....X8).
How much of this is incorrect ?
My other two questions are:
b) Is this the right way to approach this analysis in R ? Or am I better
off reading about multi-way contingency table analyses and using them ?
and
c) How do I incorporate a correction for multiple-testing into the above
analysis ? The effect of Fac on the relationship between X and Y was
planned.
I would greatly, and respectfully appreciate all pointers, tips and
admonitions.
Thank you !!!!
Suresh