[R-sig-eco] Transforming logits to probabilities
Jari Oksanen
jari.oksanen at oulu.fi
Thu Jul 8 15:41:56 CEST 2010
On 8/07/10 14:52 PM, "Manuel Spínola" <mspinola10 at gmail.com> wrote:
> Dear list members,
>
> I am using function invlogit from the "arm" package to get the
> difference in probabilities for a explanatory factor from the model
> below (I am working with factor "iluminacion").
>
Manuel,
You do not need any special packages (like "arm") to handle logits, since
basic R has function plogis() for the job.
>> invlogit(-0.4241 - 0.4804 * 0) - invlogit(-0.4241 - 0.4804 * 1) #
> difference in probability for iluminacion
> [1] 0.1074094
>
> How can I transform the standard error of the difference from logit to
> probability?
>
Make calculations in the scale of the linear predictor and then use plogis()
to change the result to the scale of response (probability). With usual
assumption.
Cheers, Jari Oksanen
> The standard error in logits is 0.2416 (taken from the model below).
>
> Best,
>
> Manuel
>
>
>> summary(mod6)
>
> Call:
> glm(formula = condicion ~ iluminacion + animales + cielo.raso +
> piso + paredes, family = binomial, data = reglog)
>
> Deviance Residuals:
> Min 1Q Median 3Q Max
> -1.2527 -1.0034 -0.8276 1.2208 1.6981
>
> Coefficients:
> Estimate Std. Error z value Pr(>|z|)
> (Intercept) -0.4241 0.2833 -1.497 0.1343
> iluminacion[T.1] -0.4804 0.2416 -1.988 0.0468 *
> animales[T.1] 0.4613 0.2385 1.934 0.0531 .
> cielo.raso[T.1] -0.1386 0.2840 -0.488 0.6254
> piso[T.1] 0.1380 0.2934 0.470 0.6382
> paredes[T.1] -0.1289 0.3012 -0.428 0.6687
> ---
> Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
>
> (Dispersion parameter for binomial family taken to be 1)
>
> Null deviance: 433.04 on 319 degrees of freedom
> Residual deviance: 424.30 on 314 degrees of freedom
> (25 observations deleted due to missingness)
> AIC: 436.3
>
> Number of Fisher Scoring iterations: 4
>
More information about the R-sig-ecology
mailing list