|z|)(Intercept) 0.522
I'll pass you the script and hypothetical data:R MASS::dose.p
dose <- c(6000, 4500, 3000, 1500, 0)
total <- c(100, 100, 100, 100, 100)
affected <- c(91, 82, 69, 49, 0)
LD SE LCL UCLSPSS
p = 0.5: 1614.444 3.207876 164.3822 15855.91
LD LCL UCL
p = 0.5: 1614.444 1198.932 1953.120
bmf2doa@gmail.com wrote:
(+ control). Our university is fond of SPSS, and I have learned to
conduct the basic probit model with it
I'll pass you the script and hypothetical data:R MASS::dose.p
dose <- c(6000, 4500, 3000, 1500, 0)
total <- c(100, 100, 100, 100, 100)
affected <- c(91, 82, 69, 49, 0)
LD SE LCL UCLSPSS
p = 0.5: 1614.444 3.207876 164.3822 15855.91
LD LCL UCL
p = 0.5: 1614.444 1198.932 1953.120
I'm too lazy to work out where your SPSS programming went wrong, but
with such a small dataset, don't you think the CIs should be wide rather
than narrow?
predict(fm1, newdata=data.frame(dose=1614.444), type="response", se.fit=TRUE) >$fit 0.5 $se.fit 0.4339901
or, using some Monte Carlo based CI estimate,
library("HelpersMG")
LD50(finney71[1:4,], equation="probit")
The LD50 is 7.387 SE 1.231
The lower limit of transitional range of doses is 5.625 SE 0.929
The higher limit of transitional range of doses is 9.149 SE 1.278
Incidentally, when we are log transforming a range of values including 0,
it is common to try and include this data point (fudge up to 0.5 or 1).
On Fri, 24 Feb 2017 23:40:55 +0000 (UTC), David Duffy
<davidD@qimr.edu.au> wrote:
bmf2doa@gmail.com wrote:
(+ control). Our university is fond of SPSS, and I have learned to
conduct the basic probit model with it
I'll pass you the script and hypothetical data:R MASS::dose.p
dose <- c(6000, 4500, 3000, 1500, 0)
total <- c(100, 100, 100, 100, 100)
affected <- c(91, 82, 69, 49, 0)
LD SE LCL UCLSPSS
p = 0.5: 1614.444 3.207876 164.3822 15855.91
LD LCL UCL
p = 0.5: 1614.444 1198.932 1953.120
Are you saying that the 0 is being ignored in both analyses, since
log(0) is undefined?
So, question: What does the "Finney71( )" mean? - beyond
I notice that the results from R are for log-dose. Results from SPSS are for dose. The results have to be different.
Results from Stata for log-dose using logistic regression (400 observations)
Coef. Std. Err. z
constant -6.893981 1.08167 -6.37
lndose .9332907 .1349154 6.92
Hi!!!have learned to conduct the basic probit model with it, including a natural logarithm transformation on my dosis data.
I'm working on the effects of alternative larvicides on Aedes aegypti. Right now, I am doing a binary mortality response with a single explanatory variable (dose) on 4 concentrations of one larvicide (+ control). Our university is fond of SPSS, and I
Not so long ago, I've started working with R, and through a combination of the 'glm' and 'dose.p' functions, I get the same slope and intercept, as well as LD50 calculations. Nevertheless, the standard errors and Z-scores calculated through the Probitmodel in SPSS comes out completely different in R. Additionally, the 95% confidence intervals for the LD50 come out differently between the two programs. I really don't have a clue on how I am getting the same slopes, intercepts and LD50's, but totally
I'll pass you the script and hypothetical data:
dose <- c(6000, 4500, 3000, 1500, 0)
total <- c(100, 100, 100, 100, 100)
affected <- c(91, 82, 69, 49, 0)
finney71 <- data.frame(dose, total, affected)
fm1 <- glm(affected/total ~ log(dose),
family=binomial(link = probit), data=finney71[finney71$dose != 0, ])
xp1 <- dose.p(fm1, p=c(0.5,0.9))
xp.ci <- xp1 + attr(xp1, "SE") %*% matrix(qnorm(1 - 0.05/2)*c(-1,1), nrow=1) EAUS.Aa <- exp(cbind(xp1, attr(xp1, "SE"), xp.ci[,1], xp.ci[,2])) dimnames(EAUS.Aa)[[2]] <- c("LD", "SE", "LCL","UCL")
So, this is the regression results I get with R:
summary(fm1)
Deviance Residuals:
1 2 3 4
0.06655 -0.02814 -0.06268 0.03474
Coefficients:
Estimate Std. Error z value
(Intercept) -6.8940 10.7802 -0.640
log(dose) 0.9333 1.3441 0.694
|z|)(Intercept) 0.522
log(dose) 0.487
(Dispersion parameter for binomial family taken to be 1)
Null deviance: 0.513878 on 3 degrees of freedom
Residual deviance: 0.010356 on 2 degrees of freedom
AIC: 6.5458
Number of Fisher Scoring iterations: 5
And the LD50 and CI transformed:
print(EAUS.Aa)
LD SE LCL UCL
p = 0.5: 1614.444 3.207876 164.3822 15855.91
p = 0.9: 6373.473 3.764879 474.1600 85669.72
These are the values I get on SPSS (just replacing the values on R output) : Coefficients:
Estimate Std. Error z value
(Intercept) -6.8940 1.082 -6.373
(dose) 2.149 0.311 6.918
And the LD50 and CI transformed:
LD LCL UCL
p = 0.5: 1614.444 1198.932 1953.120
p = 0.9: 6373.473 5145.767 9013.354
So, please if somebody can help me with this, I'd be grateful. If working with those functions won't do it, I'll use another, the one you recommend.
Thank you very much!
PD. I've already googled it but there's no satisfactory answer.
Sysop: | Keyop |
---|---|
Location: | Huddersfield, West Yorkshire, UK |
Users: | 296 |
Nodes: | 16 (2 / 14) |
Uptime: | 26:46:25 |
Calls: | 6,646 |
Calls today: | 1 |
Files: | 12,193 |
Messages: | 5,328,054 |