8
Applied Statistics Week 4 Exercise 3 Tick bites and suspicion of Borrelia Mihaela Frincu 20.12.2011

Applied Statistics Week 4 Exercise 3 Tick bites and suspicion of Borrelia Mihaela Frincu 20.12.2011

Embed Size (px)

Citation preview

Page 1: Applied Statistics Week 4 Exercise 3 Tick bites and suspicion of Borrelia Mihaela Frincu 20.12.2011

Applied Statistics

Week 4 Exercise 3

Tick bites and suspicion of Borrelia

Mihaela Frincu

20.12.2011

Page 2: Applied Statistics Week 4 Exercise 3 Tick bites and suspicion of Borrelia Mihaela Frincu 20.12.2011

Presentation of data set

gender

IgG F M

neg 1660 1252

pos 57 50

IgG – presence of Borrelia infection neg/pos

Gender – M=male, F=female

Age [years]

Page 3: Applied Statistics Week 4 Exercise 3 Tick bites and suspicion of Borrelia Mihaela Frincu 20.12.2011

1. Perform a logistic regression with IgG as response and gender as explanatory variable

fitlog<-glm(IgG~gender, family=binomial,data=borrelia)summary(fitlog)

Deviance Residuals: Min 1Q Median 3Q Max -0.2798 -0.2798 -0.2599 -0.2599 2.6097 Coefficients: Estimate Std. Error z value Pr(>|z|) (Intercept) -3.3715 0.1347 -25.028 <2e-16 ***genderM 0.1510 0.1973 0.765 0.444 ---Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 (Dispersion parameter for binomial family taken to be 1)Null deviance: 924.89 on 3018 degrees of freedomResidual deviance: 924.31 on 3017 degrees of freedomAIC: 928.31Number of Fisher Scoring iterations: 6

Page 4: Applied Statistics Week 4 Exercise 3 Tick bites and suspicion of Borrelia Mihaela Frincu 20.12.2011

2. Assess the effect of gender by a likelihood ratio test

First model: infection depends on genderfitlog<-glm(IgG~gender, family=binomial,data=borrelia)

Second model: infection is independent of gender:fitno<-glm(IgG~1, family=binomial,data=borrelia)

Comparison of the two models:anova(fitlog,fitno,test="Chisq")Analysis of Deviance TableModel 1: IgG ~ genderModel 2: IgG ~ 1 Resid. Df Resid. Dev Df Deviance Pr(>Chi)1 3017 924.31 2 3018 924.89 -1 -0.58355 0.4449

P=0.4449 there is no significant difference between the two models -> infection does not depend on gender

Page 5: Applied Statistics Week 4 Exercise 3 Tick bites and suspicion of Borrelia Mihaela Frincu 20.12.2011

3. Calculate the relative change in odds of a positive IgG test (odds ratio) due to gender

> summary(fitlog)

Coefficients: Estimate Std. Error z value Pr(>|z|) (Intercept) -3.3715 0.1347 -25.028 <2e-16 ***genderM 0.1510 0.1973 0.765 0.444 ---Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Relative change in odds ratio due to gender is:

exp(0.1510)= 1.162997

Page 6: Applied Statistics Week 4 Exercise 3 Tick bites and suspicion of Borrelia Mihaela Frincu 20.12.2011

4. Provide the odds ratio with a 95% CI

library(multcomp)

gencomp<-glht(fitlog,IgG=mcp(gender))

exp(confint(gencomp,calpha=1.96)$confint)

Estimate lwr upr

(Intercept) 0.034337350.02636945 0.04471287

genderM 1.163051400.78997917 1.71230915

Page 7: Applied Statistics Week 4 Exercise 3 Tick bites and suspicion of Borrelia Mihaela Frincu 20.12.2011

5. What value of the odds ratio corresponds to no association between gender and IgG

If infection does not depend on gender we expect the odds ratio to be 1 (equal odds).

6. Do we get the same conclusion from the 2 test, the likelihood ratio test, and the 95% CI for the odds ratio

We got the same answer.

2 test: P=0.444 there is no significant difference between the two models -> infection does not depend on gender

Likelihood ratio test: for the effect of gender p=0.4449 -> no significant gender effect

95% CI: genderM 1.16305140 0.78997917 1.71230915 ->

1 is in the confidence interval

Page 8: Applied Statistics Week 4 Exercise 3 Tick bites and suspicion of Borrelia Mihaela Frincu 20.12.2011

Reporting the results

The influence of gender on the incidence of Borrelia infections was investigated by a logistic regression with IgG as response and gender as explanatory variable. The influence of the gender on the incidence of infections was not found to be significant (p=0.444).

The dataset was also ivestigated using a likelihood ratio test. The result was similar (p=0.4449), which indicates that gender does not have a significant influence on the incidence of borrelia infection.

The odds ratio of infection male:female for the data set was found to be (95% CI) = 1.16 (0.79-1.71)

The analysis was performed in R version 2.14.0 (www.R-project.org)