R code descriptive statistics of phenotypic data by Avjinder Kaler
2 likes831 views
This document provides code examples for conducting descriptive statistics and modeling of phenotypic data in R. It covers reading data into R, calculating summary statistics like means and standard deviations, exploring data distributions through histograms and QQ plots, transforming data, fitting and comparing linear and logistic regression models, and assessing model diagnostics. The goal is to analyze phenotypic data, identify appropriate transformations and models, and check assumptions.
R code descriptive statistics of phenotypic data by Avjinder Kaler
- 1. R Code for Descriptive Statistics of Phenotypic Data Avjinder Singh Kaler
- 2. Steps 1. Reading data into R—the data file includes a header and “NA” is a missing value. data_in <‐‐‐ read.table (file="example.dat", header= T, na.string="NA") 2. Getting access to the data frame (all variables will relate to this data frame). attach(data_in) 3. Overview of data (mean, median, maximum, minimum, first and third quartiles, number of missing values). summary(data_in) 4. Calculating means and standard deviations (2 indicates that the function is applied to all columns, na.rm=T means that missing values are removed). apply(data_in,2,mean,na.rm=T) apply(data_in,2,sd,na.rm=T) 5. Means or standard deviations can be calculated for data points grouped by a factor (e.g. year). aggregate(data_in,list(data$year),FUN=mean) aggregate(data_in,list(data$year),FUN=sd) 6. Frequencies for a single variable and across two variables. table(sex) table(wormy) table(wormy,sex)
- 3. 7. Histogram. hist(FWEC) 8. xy scatter plots. plot(FWEC) 9. Box plot. boxplot(FWEC_sex,data=data_in, range=0) 10. Shapiro–Wilk’s tests to check normality of data distribution. shapiro.test(FWEC) 11. Checking data distribution with QQ plot—if data are normally distributed, the plotted data and the line are well aligned. qqplot(FWEC) qqline(FWEC) 12. Data transformation—log, square root, and cube root transformation. log_FWEC <‐‐‐ log(FWEC) sqrt_FWEC <‐‐‐ sqrt(FWEC+1) cbrt_FWEC <‐‐‐ (FWEC)^(1/3)
- 4. 13. Box–Cox transformation. #CodetofindsuitablelambdaforYtothepowerlambda #download thelibrary(MASS) #seq(min value, max value, step) defines the range from which lambda is drawn boxcox(FWEC_factor(sex)+factor(birth_rearing_ type), lambda = seq(0,1.0,0.01) savePlot("boxcox","jpeg") lambda = "insert maximum lambda value in graph here" trans(FWEC) <- ((FWEC^lambda)-1)/lambda MASS library 14. Checking homogeneity of variances. #download library (Rcmdr) library(Rcmdr) #run the Leven’s test, specifying the vector of data y and group, the factor across which the variances are tested (e.g., year) leveneTest(y,group) 15. Fitting a linear model and ANOVA. #need to load the "car" package for Type III ANOVA library(car)
- 5. lmod <- lm(cbrt_FWEC_factor(sex)) #Type I ANOVA anova(lm) #Type III ANOVA---Note that the first letter in the commandbelow has to be a capital "A" (ensure that you loaded the "car" package as shown above) Anova(lmod, type¼"III") 16. Addressing confounding of explanatory variables in a linear model. lmod1 <- lm(cbrt_FWEC_factor(sex)+factor(birth_ type)*factor(rearing_type)) lmod2 <-lm(cbrt_FWEC_factor(sex)+factor(birth_ rearing_type)) 17. Check the difference with an ANOVA. Anova(lmod1,type="III") Anova(lmod2,type="III") 18. Model comparison using logistic regression for binary data. logres <- glm(formula=wormy_status_factor(sex) + factor(birth_rearing_type), family = binomial (link="logit"))
- 6. #producing an analysis-of-deviance table to test fixed effects anova(logres,test="Chisq") #produces the deviance of the model (the lower the better the fit) summary(glm(formula=wormy_status_factor(sex) + factor(birth_rearing_type), family = binomial (link="logit"))$deviance)) #the difference in deviance can be formally tested with a loglikelihood ratio test #install library(lme4) library(lme4) #comparing two nested models ("nested" means that one has one more factor than the other) logres1 <- lmer(wormy_status_factor(sex)), family = "binomial", method="Laplace") logres2 <- lmer(wormy_status_factor(sex) + factor (birth_rearing_type), family = "binomial", Method="Laplace") anova(logres1,logres2) #to assess the model, plot predicted probability against observed proportion #install library(languageR) library(languageR) plot.logistics.fit.fnc(logres1,logres2)
- 7. 19. Model diagnostics. #the following produces plot of residual vs. fitted value, QQ plot, and scale-location plot of the previously tested model 1 (lmod1) plot(lmod1) #assessing a logit model for binary data by plotting the predicted probability against observed proportions #download library(languageR) library(languageR) plot.logistic.fit.fnc(logres1,data_in) 20. Extracting residuals and writing them to a file—assuming lmod2 is the model of choice. res_lmod2 <-residuals(lmod2) write.table(res_lmod2,file¼"res_FWEC")