110 likes | 265 Views
Stat 6601 project Linear Statistical Models Analysis of Covariance Example. By Gadir Marian Myrna Moreno. Data.
E N D
Stat 6601 project Linear Statistical ModelsAnalysis of Covariance Example By Gadir Marian Myrna Moreno
Data • ‘Whiteside’ data, Mr. Derek recorded weekly gas consumption and average external temperature at his house during two ‘heating seasons’ one before and after cavity-wall insulation was installed. • Variables: - Insul (levels: before or after insulation) - Temp (the average outside temperature in degrees Celsius) -Gas (The weekly gas consumption in 1000 cubic feet units)
Goal • Assess the effect of the insulation on gas consumption.
Method • Linear Model for Analysis of Covariance Y= +X + Where: is a random effect due to treatment. is a fixed effect due to covariate. is a random error.
Method(continued) • Using R: -A primary model is fitted using a “model fitting function” lm (formula, data, weights, subset, na.action) - A resulting “fitted model object” can be analysed, interrogated or modified.
Codes • require(latice) • xyplot(Gas ~ Temp | Insul, whiteside, panel = • function(x, y, ...) { • panel.xyplot(x, y, ...) • panel.lmline(x, y, ...) • }, xlab = "Average external temperature (deg. C)", • ylab = "Gas consumption (1000 cubic feet)", aspect = "xy", • strip = function(...) strip.default(..., style = 1)) • gasB <- lm(Gas ~ Temp, whiteside, subset = Insul=="Before") • gasA <- update(gasB, subset = Insul=="After") • summary(gasB) • summary(gasA) • gasBA <- lm(Gas ~ Insul/Temp - 1, whiteside) • summary(gasBA) • gasQ <- lm(Gas ~ Insul/(Temp + I(Temp^2)) - 1, whiteside) summary(gasQ)$coef • gasPR <- lm(Gas ~ Insul + Temp, whiteside) • anova(gasPR, gasBA) • options(contrasts = c("contr.treatment", "contr.poly")) • gasBA1 <- lm(Gas ~ Insul*Temp, whiteside) • summary(gasBA1)$coef
Results • The output from fitting regression model: Residuals: Min 1Q Median 3Q Max -0.97802 -0.18011 0.03757 0.20930 0.63803 Coefficients: Estimate Std. Error t value Pr(>|t|) InsulBefore 6.85383 0.13596 50.41 <2e-16 *** InsulAfter 4.72385 0.11810 40.00 <2e-16 *** InsulBefore:Temp -0.39324 0.02249 -17.49 <2e-16 *** InsulAfter:Temp -0.27793 0.02292 -12.12 <2e-16 *** Residual standard error: 0.323 on 52 degrees of freedom
Results(continued) The output by fitting quadratic regression model: Estimate Std. Error t value Pr(>|t|) InsulBefore 6.759215179 0.150786777 44.826312 4.854615e-42 InsulAfter 4.496373920 0.160667904 27.985514 3.302572e-32 InsulBefore:Temp -0.31765873 0.062965170 -5.044991 6.362323e-06 InsulAfter:Temp -0.137901603 0.073058019 -1.887563 6.489554e-02 InsulBefore:I(Temp^2) -0.008472572 0.006624737 -1.278930 2.068259e-01 InsulAfter:I(Temp^2) -0.014979455 0.007447107 -2.011446 4.968398e-02
Results(continued) The output from the ANOVA Estimate Std. Error t value Pr(>|t|) (Intercept) 6.8538277 0.13596397 50.409146 7.997414e-46 InsulAfter -2.1299780 0.18009172 -11.827185 2.315921e-16 Temp -0.3932388 0.02248703 -17.487358 1.976009e-23 InsulAfter:Temp 0.1153039 0.03211212 3.590665 7.306852e-04
Summary • Whiteside data • Fitting Linear Regression Model • Fitting Quadratic Regression Model