On Feb 1, 2010, at 5:01 PM, Amy Mikhail wrote:
> Dear David,
>
> Many thanks for your reply. My rapid test results are categorical
> (positive negative) and I have converted them to factors. My gold
> standard (parasitaemia on a blood slide) is a continuous numerical
> variable but can be categorical too.
>
> I tried xtabs, but it doesn't seem to like the use of formulas with
> my input (cp.r is factoral, r2.vc is continuous numerical) :
>
I did not suggest using xtabs with a continuous variable.
> ref <- ftable(xtabs(cp.r ~ r2.vc, data = res))
>
> Error in Summary.factor(c(2L, 3L, 2L, 3L, 3L, 3L, 2L, 2L, 2L, 2L,
> 2L, :
> sum not meaningful for factors
>
> I have tried chisq.test and fisher.test too, they work fine but the
> input is somewhat more laborious than I would like - I was hoping
> there might be a wrapper function that would do two things (a)
> create a 2x2 table and (b) then do fisher or chi square on it (so
> that the input are the actual variables)...
> set.seed(7)
> Paralev<- sample(c("Low", "High"), 250, replace=TRUE)
> Result <- sample(c("False neg", "True Pos"), 250, prob=c(1/250,
249/250), replace=TRUE)
> xtabs(data=cbind(Paralev, Result))
Result
Paralev False neg True Pos
High 1 126
Low 0 123
> fisher.test( xtabs(data=cbind(Paralev, Result)) )
Fisher's Exact Test for Count Data
data: xtabs(data = cbind(Paralev, Result))
p-value = 1
alternative hypothesis: true odds ratio is not equal to 1
95 percent confidence interval:
0.02483 Inf
sample estimates:
odds ratio
Inf
>
> If I create a matrix and then do fisher's on it, the result is as I
> expect:
>
> > ParaSens <- matrix(c(1,0,124,130), nrow = 2, dimnames =
> list(Paralev = c("Low", "High"), Result = c("False negative", "True
> positive")))
> > ParaSens
> Result
> Paralev False negative True positive
> Low 1 124
> High 0 130
> > fisher.test(ParaSens, alternative = "greater")
>
> Fisher's Exact Test for Count Data
>
> data: ParaSens
> p-value = 0.4902
> alternative hypothesis: true odds ratio is greater than 1
> 95 percent confidence interval:
> 0.05473174 Inf
> sample estimates:
> odds ratio
> Inf
>
>
> If I try to use the stacktable function in epicalc, however (because
> I have 6 rdts
What is a "rdts"?
> to test), it gives me the wrong result (cp.r - tested separately
> above - should not be significantly different between the two levels
> of parasitaemia):
This is gobbleygook to me. Consider providing
dput(<a name of an R object>)
>
> tableStack(vars=6:11, by=paralev)
> negative < 5,000 T/uL > 5,000 T/uL Test
> stat. P value
> cp.r Fisher's exact test
> < 0.001
> Invalid 1 (3.7) 6 (4.6) 4 (3)
> Negative 22 (81.5) 1 (0.8) 0 (0)
> Pv 4 (14.8) 124 (94.7) 130 (97)
>
> cp.c Fisher's exact test
> < 0.001
> Invalid 1 (3.7) 3 (2.3) 2 (1.5)
> Negative 22 (81.5) 1 (0.8) 0 (0)
> Pv 4 (14.8) 127 (96.9) 132 (98.5)
>
> cv.r Fisher's exact test
> < 0.001
> Invalid 0 (0) 3 (2.3) 1 (0.7)
> Negative 23 (85.2) 19 (14.5) 1 (0.7)
> Pv 4 (14.8) 109 (83.2) 132 (98.5)
>
> cv.c Fisher's exact test
> < 0.001
> Invalid 0 (0) 3 (2.3) 3 (2.2)
> Negative 23 (85.2) 18 (13.7) 1 (0.7)
> Pv 4 (14.8) 110 (84) 130 (97)
>
> sv.r Fisher's exact test
> < 0.001
> Invalid 5 (18.5) 14 (10.7) 31 (23.1)
> Mixed 0 (0) 0 (0) 1 (0.7)
> Negative 22 (81.5) 7 (5.3) 0 (0)
> Pv 0 (0) 110 (84) 102 (76.1)
>
> sv.c Fisher's exact test
> < 0.001
> Invalid 4 (14.8) 8 (6.1) 19 (14.2)
> Negative 20 (74.1) 5 (3.8) 0 (0)
> Pv 3 (11.1) 118 (90.1) 115 (85.8)
>
> Please note that in tableStack I have a third (unwanted) factor
> level (Invalid): maybe this is what is causing the problem, since
> the above is a 3x2 table wherease when I create a matrix separately
> I leave out the invalids so that it is only a 2x2 table?
>
> Regarding your comment on continuous variables - had not really come
> across that one before, for malaria rapid diagnostic tests (which I
> am dealing with) the sensitivity and specificity is always reported
> with reference to either microscopy results or PCR results as the
> gold standard. PPV has some value too in that it can take into
> account invalid RDT results, wherease the other two measurements
> can't.
>
> Best,
> Amy
>
>
snipped earlier material.
David Winsemius, MD
Heritage Laboratories
West Hartford, CT
[[alternative HTML version deleted]]
_______________________________________________
R-sig-Epi@stat.math.ethz.ch mailing list
https://stat.ethz.ch/mailman/listinfo/r-sig-epi
