class: center, middle, inverse, title-slide # STA 326 2.0 Programming and Data Analysis with R ## ✍️ Regression Analysis ### ### Dr Thiyanga Talagala --- <style type="text/css"> .remark-slide-content { font-size: 35px; } </style> <style type="text/css"> h1, #TOC>ul>li { color: #006837; font-weight: bold; } h2, #TOC>ul>ul>li { color: #006837; #font-family: "Times"; font-weight: bold; } h3, #TOC>ul>ul>li { color: #ce1256; #font-family: "Times"; font-weight: bold; } </style> .pull-left[ ## Today's menu - Fit a model - Visualise the fitted model - Measure the strength of the fit - Residual analysis - Make predictions ] .pull-right[ <center><img src="cakemodel.jpeg" height="600px"/></center> ] --- background-image: url('box.jpeg') background-position: center background-size: contain --- # Packages ```r library(broom) library(modelr) library(GGally) library(carData) library(tidyverse) library(magrittr) library(car) # to calculate VIF ``` --- # Data: Prestige of Canadian Occupations ```r head(Prestige, 5) ``` ``` education income women prestige census type gov.administrators 13.11 12351 11.16 68.8 1113 prof general.managers 12.26 25879 4.02 69.1 1130 prof accountants 12.77 9271 15.70 63.4 1171 prof purchasing.officers 11.42 8865 9.11 56.8 1175 prof chemists 14.62 8403 11.68 73.5 2111 prof ``` --- ```r summary(Prestige) ``` ``` education income women prestige Min. : 6.380 Min. : 611 Min. : 0.000 Min. :14.80 1st Qu.: 8.445 1st Qu.: 4106 1st Qu.: 3.592 1st Qu.:35.23 Median :10.540 Median : 5930 Median :13.600 Median :43.60 Mean :10.738 Mean : 6798 Mean :28.979 Mean :46.83 3rd Qu.:12.648 3rd Qu.: 8187 3rd Qu.:52.203 3rd Qu.:59.27 Max. :15.970 Max. :25879 Max. :97.510 Max. :87.20 census type Min. :1113 bc :44 1st Qu.:3120 prof:31 Median :5135 wc :23 Mean :5402 NA's: 4 3rd Qu.:8312 Max. :9517 ``` --- # Data description **`prestige`**: prestige of Canadian occupations, measured by the Pineo-Porter prestige score for occupation taken from a social survey conducted in the mid-1960s. **`education`**: Average education of occupational incumbents, years, in 1971. **`income`**: Average income of incumbents, dollars, in 1971. **`women`**: Percentage of incumbents who are women. --- # Data description (cont.) **`census`**: Canadian Census occupational code. **`type`**: Type of occupation. - prof: professional and technical - wc: white collar - bc: blue collar - NA: missing The dataset consists of 102 observations, each corresponding to a particular occupation. --- # Training test and Test set ```r ## Create an ID per row df <- Prestige %>% mutate(id = row_number()) ## set the seed to make your partition reproducible set.seed(123) train <- df %>% sample_frac(.80) dim(train) ``` ``` [1] 82 7 ``` ```r test <- anti_join(df, train, by = 'id') dim(test) ``` ``` [1] 20 7 ``` --- # Exploratory Data Analysis ```r train ``` ``` education income women prestige census type id medical.technicians 12.79 5180 76.04 67.5 3156 wc 31 welders 7.92 6477 5.17 41.8 8335 bc 79 commercial.travellers 11.13 8780 3.16 40.2 5133 wc 51 economists 14.44 8049 57.31 62.2 2311 prof 14 farmers 6.84 3643 3.60 44.1 7112 <NA> 67 receptionsts 11.04 2901 92.86 38.7 4171 wc 42 sales.supervisors 9.84 7482 17.04 41.5 5130 wc 50 mail.carriers 9.22 5511 7.62 36.1 4172 wc 43 taxi.drivers 7.93 4224 3.59 25.1 9173 bc 99 veterinarians 15.94 14558 4.32 66.7 3115 prof 25 construction.foremen 8.24 8880 0.65 51.1 8780 bc 90 carpenters 6.92 5299 0.56 38.9 8781 bc 91 rotary.well.drillers 8.88 6860 0.00 35.3 7711 bc 69 buyers 11.03 7956 23.88 51.1 5191 wc 57 civil.engineers 14.52 11377 1.03 73.1 2143 prof 9 slaughterers.2 7.64 5134 17.26 34.8 8215 bc 72 osteopaths.chiropractors 14.71 17498 6.91 68.4 3117 prof 26 biologists 15.09 8258 25.65 72.6 2133 prof 7 train.engineers 8.49 8845 0.00 48.9 9131 bc 97 electrical.linemen 9.05 8316 1.34 40.9 8731 bc 88 typists 11.49 3148 95.97 41.9 4113 wc 36 sheet.metal.workers 8.40 6565 2.30 35.9 8333 bc 78 construction.labourers 7.52 3910 1.09 26.5 8798 bc 95 tool.die.makers 10.09 8043 1.50 42.5 8311 bc 76 psychologists 14.36 7405 48.28 74.9 2315 prof 15 commercial.artists 11.09 6197 21.03 57.2 3314 prof 32 auto.repairmen 8.10 5795 0.81 38.1 8581 bc 85 radio.tv.repairmen 10.29 5449 2.92 37.2 8537 bc 83 file.clerks 12.09 3016 83.19 32.7 4161 wc 41 secondary.school.teachers 15.08 8034 46.80 66.1 2733 prof 23 nurses 12.46 4614 96.12 64.7 3131 prof 27 cooks 7.74 3116 52.00 29.7 6121 bc 60 newsboys 9.62 918 7.00 14.8 5143 <NA> 53 typesetters 10.00 6462 13.58 42.2 9511 bc 101 bakers 7.54 4199 33.30 38.9 8213 bc 70 railway.sectionmen 6.67 4696 0.00 27.3 8715 bc 87 tellers.cashiers 10.64 2448 91.76 42.3 4133 wc 38 athletes 11.44 8206 8.13 54.1 3373 <NA> 34 physio.therapsts 13.62 5092 82.66 72.1 3137 prof 29 chemists 14.62 8403 11.68 73.5 2111 prof 5 architects 15.44 14163 2.69 78.1 2141 prof 8 draughtsmen 12.30 7059 7.83 60.0 2163 prof 12 computer.programers 13.83 8425 15.33 53.8 2183 prof 13 librarians 14.15 6112 77.10 58.1 2351 prof 18 radio.tv.announcers 12.71 7562 11.15 57.6 3337 wc 33 electricians 9.93 7147 0.99 50.2 8733 bc 89 bus.drivers 7.58 5562 9.47 35.9 9171 bc 98 house.painters 7.81 4549 2.46 29.9 8785 bc 93 elevator.operators 7.58 3582 30.08 20.1 6193 bc 66 university.teachers 15.97 12480 19.59 84.6 2711 prof 21 aircraft.workers 8.78 6573 5.78 43.7 8515 bc 81 textile.weavers 6.69 4443 31.36 33.3 8267 bc 74 claim.adjustors 11.13 5052 56.10 51.1 4192 wc 47 aircraft.repairmen 10.10 7716 0.78 50.3 8582 bc 86 bookbinders 8.55 3617 70.87 35.2 9517 bc 102 social.workers 14.21 6336 54.77 55.1 2331 prof 16 pharmacists 15.21 10432 24.71 69.3 3151 prof 30 physicists 15.64 11030 5.13 77.6 2113 prof 6 auto.workers 8.43 5811 13.62 35.9 8513 bc 80 funeral.directors 10.57 7869 6.01 54.9 6141 bc 62 primary.school.teachers 13.62 5648 83.78 59.6 2731 prof 22 sewing.mach.operators 6.38 2847 90.67 28.2 8563 bc 84 computer.operators 11.36 4330 75.92 47.7 4143 wc 39 travel.clerks 11.43 6259 39.17 35.7 4193 wc 48 lawyers 15.77 19263 5.13 82.3 2343 prof 17 janitors 7.11 3472 33.57 17.3 6191 bc 65 purchasing.officers 11.42 8865 9.11 56.8 1175 prof 4 slaughterers.1 7.64 5134 17.26 25.2 8215 bc 71 babysitters 9.46 611 96.53 25.9 6147 <NA> 63 insurance.agents 11.60 8131 13.09 47.3 5171 wc 55 ministers 14.50 4686 4.14 72.8 2511 prof 20 longshoremen 8.37 4753 0.00 26.1 9313 bc 100 firefighters 9.47 8895 0.00 43.5 6111 bc 58 plumbers 8.33 6928 0.61 42.9 8791 bc 94 collectors 11.20 4741 47.06 29.4 4191 wc 46 canners 7.42 1890 72.24 23.2 8221 bc 73 accountants 12.77 9271 15.70 63.4 1171 prof 3 postal.clerks 10.07 3739 52.27 37.2 4173 wc 44 pilots 12.27 14032 0.58 66.1 9111 prof 96 bartenders 8.50 3930 15.51 20.2 6123 bc 61 launderers 7.33 3000 69.31 20.8 6162 bc 64 office.clerks 11.00 4075 63.23 35.6 4197 wc 49 ``` ] --- ```r Prestige_1 <- train %>% pivot_longer(c(1, 2, 3, 4), names_to="variable", values_to="value") Prestige_1 ``` ``` # A tibble: 328 x 5 census type id variable value <int> <fct> <int> <chr> <dbl> 1 3156 wc 31 education 12.8 2 3156 wc 31 income 5180 3 3156 wc 31 women 76.0 4 3156 wc 31 prestige 67.5 5 8335 bc 79 education 7.92 6 8335 bc 79 income 6477 7 8335 bc 79 women 5.17 8 8335 bc 79 prestige 41.8 9 5133 wc 51 education 11.1 10 5133 wc 51 income 8780 # … with 318 more rows ``` --- ```r ggplot(Prestige_1, aes(x=value)) + geom_histogram() + facet_wrap(variable ~., ncol=1) ``` <img src="l132021_files/figure-html/unnamed-chunk-8-1.png" height="80%" /> --- ```r ggplot(Prestige_1, aes(x=value)) + geom_histogram() + facet_wrap(variable ~., ncol=1, scales = "free") ``` <img src="l132021_files/figure-html/unnamed-chunk-9-1.png" height="80%" /> --- ```r ggplot(Prestige_1, aes(x=value)) + geom_histogram(colour="white") + facet_wrap(variable ~., ncol=1, scales = "free") ``` <img src="l132021_files/figure-html/unnamed-chunk-10-1.png" height="80%" /> --- ```r ggplot(Prestige_1, aes(x=value, fill=variable)) + geom_density() + facet_wrap(variable ~., ncol=1, scales = "free") ``` <img src="l132021_files/figure-html/unnamed-chunk-11-1.png" height="80%" /> --- ```r ggplot(Prestige_1, aes(y = value, x = type, fill = type)) + geom_boxplot() + facet_wrap(variable ~., ncol=1, scales = "free") ``` <img src="l132021_files/figure-html/unnamed-chunk-12-1.png" height="80%" /> --- ```r *Prestige_1 %>% * filter(is.na(type) == FALSE) %>% ggplot(aes(y=value, x=type, fill=type)) + geom_boxplot() + facet_wrap(variable ~., ncol=1, scales = "free") ``` <img src="l132021_files/figure-html/unnamed-chunk-13-1.png" height="70%" /> --- ```r Prestige_1 %>% filter(is.na(type) == FALSE) %>% ggplot(aes(x = value, y = type, fill = type)) + geom_boxplot() + facet_wrap(variable ~., ncol=1, scales = "free") ``` <img src="l132021_files/figure-html/unnamed-chunk-14-1.png" height="70%" /> --- ```r train %>% select(education, income, prestige, women) %>% ggpairs() ``` <!-- --> --- ```r train %>% filter(is.na(type) == FALSE) %>% ggpairs(columns= c("education", "income", "prestige", "women"), mapping=aes(color=type)) ``` <!-- --> --- class: duke-orange, center, middle # Regression analysis --- # Steps 1. Fit a model. 2. Visualize the fitted model. 3. Measuring the strength of the fit. 4. Residual analysis. 5. Interpret the coefficients. 6. Make predictions using the fitted model. --- class: duke-orange, middle # Model 1: prestige ~ education --- ## Recap **True relationship between X and Y in the population** `$$Y = f(X) + \epsilon$$` **If `\(f\)` is approximated by a linear function** `$$Y = \beta_0 + \beta_1X + \epsilon$$` The error terms are normally distributed with mean `\(0\)` and variance `\(\sigma^2\)`. Then the mean response, `\(Y\)`, at any value of the `\(X\)` is `$$E(Y|X=x_i) = E(\beta_0 + \beta_1x_i + \epsilon)=\beta_0+\beta_1x_i$$` --- For a single unit `\((y_i, x_i)\)` `$$y_i = \beta_0 + \beta_1x_i+\epsilon_i \text{ where } \epsilon_i \sim N(0, \sigma^2)$$` We use sample values `\((y_i, x_i)\)` where `\(i=1, 2, ...n\)` to estimate `\(\beta_0\)` and `\(\beta_1\)`. The fitted regression model is `$$\hat{Y_i} = \hat{\beta}_0 + \hat{\beta}_1x_i$$` --- # How to estimate `\(\beta_0\)` and `\(\beta_1\)`? Sum of squares of Residuals `$$SSR=e_1^2+e_2^2+...+e_n^2$$` The least-squares regression approach chooses coefficients `\(\hat{\beta}_0\)` and `\(\hat{\beta}_1\)` to minimize `\(SSR\)`. --- background-image: url('reg/reg1.PNG') background-position: center background-size: contain --- background-image: url('reg/reg2.PNG') background-position: center background-size: contain --- background-image: url('reg/reg3.PNG') background-position: center background-size: contain --- background-image: url('reg/reg4.PNG') background-position: center background-size: contain --- background-image: url('reg/reg5.PNG') background-position: center background-size: contain --- background-image: url('reg/reg6.PNG') background-position: center background-size: contain --- background-image: url('reg/reg7.PNG') background-position: center background-size: contain --- background-image: url('reg/errors.PNG') background-position: center background-size: contain --- class: duke-orange, middle # Model 1: prestige ~ education ### 1. Fit a model --- # Model 1: Fit a model `$$y_i = \beta_0 + \beta_1x_i + \epsilon_i, \text{where } \epsilon_i \sim NID(0, \sigma^2)$$` To estimate `\(\beta_0\)` and `\(\beta_1\)` ```r model1 <- lm(prestige ~ education, data=train) ``` --- ```r summary(model1) ``` ``` Call: lm(formula = prestige ~ education, data = train) Residuals: Min 1Q Median 3Q Max -26.1683 -6.1403 0.8302 6.2665 17.8892 Coefficients: Estimate Std. Error t value Pr(>|t|) (Intercept) -10.0990 3.8922 -2.595 0.0113 * education 5.3085 0.3519 15.085 <2e-16 *** --- Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1 Residual standard error: 8.81 on 80 degrees of freedom Multiple R-squared: 0.7399, Adjusted R-squared: 0.7366 F-statistic: 227.5 on 1 and 80 DF, p-value: < 2.2e-16 ``` --- # What's messy about the output? ``` Call: lm(formula = prestige ~ education, data = train) Residuals: Min 1Q Median 3Q Max -26.1683 -6.1403 0.8302 6.2665 17.8892 Coefficients: Estimate Std. Error t value Pr(>|t|) (Intercept) -10.0990 3.8922 -2.595 0.0113 * education 5.3085 0.3519 15.085 <2e-16 *** --- Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1 Residual standard error: 8.81 on 80 degrees of freedom Multiple R-squared: 0.7399, Adjusted R-squared: 0.7366 F-statistic: 227.5 on 1 and 80 DF, p-value: < 2.2e-16 ``` --- - Extract coefficients takes multiple steps. ```r data.frame(coef(summary(model1))) ``` - Column names are inconvenient to use. - Information are stored in row names. --- # `broom` functions - broom takes model objects and turns them into tidy data frames that can be used with other tidy tools. - Three main functions `tidy()`: component-level statistics `augment()`: observation-level statistics `glance()`: model-level statistics --- # Component-level statistics: `tidy()` ```r model1 %>% tidy() ``` ``` # A tibble: 2 x 5 term estimate std.error statistic p.value <chr> <dbl> <dbl> <dbl> <dbl> 1 (Intercept) -10.1 3.89 -2.59 1.13e- 2 2 education 5.31 0.352 15.1 4.17e-25 ``` ```r model1 %>% tidy(conf.int=TRUE) ``` ``` # A tibble: 2 x 7 term estimate std.error statistic p.value conf.low conf.high <chr> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> 1 (Intercept) -10.1 3.89 -2.59 1.13e- 2 -17.8 -2.35 2 education 5.31 0.352 15.1 4.17e-25 4.61 6.01 ``` --- # Component-level statistics: `tidy()` ```r model1 %>% tidy() %>% select(term, estimate) ``` ``` # A tibble: 2 x 2 term estimate <chr> <dbl> 1 (Intercept) -10.1 2 education 5.31 ``` --- # Component-level statistics: `tidy()` (cont.) ```r model1 %>% tidy() ``` ``` # A tibble: 2 x 5 term estimate std.error statistic p.value <chr> <dbl> <dbl> <dbl> <dbl> 1 (Intercept) -10.1 3.89 -2.59 1.13e- 2 2 education 5.31 0.352 15.1 4.17e-25 ``` Fitted model is `$$\hat{Y}_i = -9.42 + 5.27 x_i$$` --- # Why are tidy model outputs useful? ```r tidy_model1 <- model1 %>% tidy(conf.int=TRUE) ggplot(tidy_model1, aes(x=estimate, y=term, color=term)) + geom_point() + geom_errorbarh(aes(xmin = conf.low, xmax=conf.high))+ggtitle("Coefficient plot") ``` <!-- --> --- class: duke-orange, middle # Model 1: prestige ~ education ### 1. Fit a model ### 2. Visualise the fitted model --- # Model 1: Visualise the fitted model <!-- --> --- ## Model 1: Visualise the fitted model (style the line) ```r ggplot(data=train, aes(y=prestige, x=education)) + geom_point(alpha=0.5) + geom_smooth(method="lm", se=FALSE, * col="forestgreen", lwd=2) ``` <!-- --> --- class: duke-orange, middle # Model 1: prestige ~ education ### 1. Fit a model ### 2. Visualise the fitted model ### 3. Measure the strength of the fit --- # Model-level statistics: `glance()` Measuring the strength of the fit ```r glance(model1) ``` ``` # A tibble: 1 x 12 r.squared adj.r.squared sigma statistic p.value df logLik AIC BIC <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> 1 0.740 0.737 8.81 228. 4.17e-25 1 -294. 594. 601. # … with 3 more variables: deviance <dbl>, df.residual <int>, nobs <int> ``` ```r glance(model1)$adj.r.squared # extract values ``` ``` [1] 0.7366237 ``` Roughly 73% of the variability in prestige can be explained by the variable education. --- class: duke-orange, middle # Model 1: prestige ~ education ### 1. Fit a model ### 2. Visualise the fitted model ### 3. Measure the strength of the fit ### 4. Residual analysis --- # Observation-level statistics: `augment()` ```r model1_fitresid <- augment(model1) model1_fitresid ``` ``` # A tibble: 82 x 9 .rownames prestige education .fitted .resid .hat .sigma .cooksd .std.resid <chr> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> 1 medical.t… 67.5 12.8 57.8 9.70 0.0191 8.80 1.20e-2 1.11 2 welders 41.8 7.92 31.9 9.86 0.0246 8.79 1.62e-2 1.13 3 commercia… 40.2 11.1 49.0 -8.78 0.0125 8.81 6.36e-3 -1.00 4 economists 62.2 14.4 66.6 -4.36 0.0344 8.85 4.51e-3 -0.503 5 farmers 44.1 6.84 26.2 17.9 0.0361 8.63 8.01e-2 2.07 6 reception… 38.7 11.0 48.5 -9.81 0.0124 8.80 7.86e-3 -1.12 7 sales.sup… 41.5 9.84 42.1 -0.636 0.0134 8.87 3.59e-5 -0.0727 8 mail.carr… 36.1 9.22 38.8 -2.74 0.0157 8.86 7.88e-4 -0.314 9 taxi.driv… 25.1 7.93 32.0 -6.90 0.0245 8.83 7.90e-3 -0.793 10 veterinar… 66.7 15.9 74.5 -7.82 0.0559 8.82 2.47e-2 -0.913 # … with 72 more rows ``` --- # Residuals and Fitted Values <!-- --> --- # Residuals and Fitted Values .pull-left[ <!-- --> ] .pull-right[ The residual is the difference between the observed and predicted response. The residual for the `\(i^{th}\)` observation is `$$e_i = y_i - \hat{Y}_i=y_i - (\hat{\beta_0}+\hat{\beta_1}x_i)$$` ] --- ## Conditions for inference for regression - The relationship between the response and the predictors is linear. - The error terms are assumed to have zero mean and unknown constant variance `\(\sigma^2\)`. - The errors are normally distributed. - The errors are uncorrelated. --- # Plot of residuals in time sequence. - The errors are uncorrelated. - Often, we can conclude that the this assumption is sufficiently met based on a description of the data and how it was collected. --- background-image: url('reg/errors.PNG') background-position: right background-size: contain ### Plot of residuals vs fitted values .pull-left[ This is useful for detecting several common types of model inadequacies. ] --- ## Plot of residuals vs fitted values and Plot of residuals vs predictor ### linearity and constant variance .pull-left[ Residuals vs Fitted ```r ggplot(model1_fitresid, aes(x = .fitted, y = .resid))+ ------ + ------ ``` ] .pull-right[ Residuals vs X ```r ggplot(model1_fitresid, aes(x = education, y = .resid))+ ------ + ------ ``` ] --- .pull-left[ Residuals vs Fitted <!-- --> ] .pull-right[ Residuals vs X <!-- --> ] --- # Normality of residuals .pull-left[ ```r ggplot(model1_fitresid, aes(x=.resid))+ geom_histogram(colour="white")+ggtitle("Distribution of Residuals") ``` <!-- --> ] .pull-right[ ```r ggplot(model1_fitresid, aes(sample=.resid))+ stat_qq() + stat_qq_line()+labs(x="Theoretical Quantiles", y="Sample Quantiles") ``` <!-- --> ] --- ```r shapiro.test(model1_fitresid$.resid) ``` ``` Shapiro-Wilk normality test data: model1_fitresid$.resid W = 0.97547, p-value = 0.1176 ``` --- class: duke-orange, middle # Model 2: prestige ~ education + `income` ### 1. Fit a model ### 2. Visualise the fitted model ### 3. Measure the strength of the fit ### 4. Residual analysis --- ### Model 2: prestige ~ education + `income` ```r model2 <- lm(prestige ~ income + education, data=train) summary(model2) ``` ``` Call: lm(formula = prestige ~ income + education, data = train) Residuals: Min 1Q Median 3Q Max -18.4585 -5.1455 0.0464 5.0861 18.0320 Coefficients: Estimate Std. Error t value Pr(>|t|) (Intercept) -7.9953726 3.4549442 -2.314 0.0233 * income 0.0015822 0.0003222 4.911 4.79e-06 *** education 4.1373699 0.3910785 10.579 < 2e-16 *** --- Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1 Residual standard error: 7.76 on 79 degrees of freedom Multiple R-squared: 0.8007, Adjusted R-squared: 0.7957 F-statistic: 158.7 on 2 and 79 DF, p-value: < 2.2e-16 ``` --- # Model 2: prestige ~ education + `income` (cont.) ```r model2_fitresid <- augment(model2) model2_fitresid ``` ``` # A tibble: 82 x 10 .rownames prestige income education .fitted .resid .hat .sigma .cooksd <chr> <dbl> <int> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> 1 medical.techn… 67.5 5180 12.8 53.1 14.4 0.0342 7.63 4.20e-2 2 welders 41.8 6477 7.92 35.0 6.78 0.0311 7.77 8.44e-3 3 commercial.tr… 40.2 8780 11.1 51.9 -11.7 0.0185 7.69 1.47e-2 4 economists 62.2 8049 14.4 64.5 -2.28 0.0374 7.80 1.16e-3 5 farmers 44.1 3643 6.84 26.1 18.0 0.0361 7.53 6.99e-2 6 receptionsts 38.7 2901 11.0 42.3 -3.57 0.0391 7.80 2.99e-3 7 sales.supervi… 41.5 7482 9.84 44.6 -3.05 0.0174 7.80 9.32e-4 8 mail.carriers 36.1 5511 9.22 38.9 -2.77 0.0157 7.80 6.90e-4 9 taxi.drivers 25.1 4224 7.93 31.5 -6.40 0.0247 7.77 5.88e-3 10 veterinarians 66.7 14558 15.9 81.0 -14.3 0.0847 7.62 1.14e-1 # … with 72 more rows, and 1 more variable: .std.resid <dbl> ``` --- ## Plot of residuals vs fitted values .pull-left[ <!-- --> ] .pull-right[ linearity and constant variance? ] --- # Normality of residuals .pull-left[ ```r ggplot(model2_fitresid, aes(x=.resid))+ geom_histogram(colour="white")+ggtitle("Distribution of Residuals") ``` <!-- --> ] .pull-right[ ```r ggplot(model2_fitresid, aes(sample=.resid))+ stat_qq() + stat_qq_line()+labs(x="Theoretical Quantiles", y="Sample Quantiles") ``` <!-- --> ] --- ```r shapiro.test(model2_fitresid$.resid) ``` ``` Shapiro-Wilk normality test data: model2_fitresid$.resid W = 0.99298, p-value = 0.9405 ``` --- ## Plot of residuals vs predictor variables .pull-left[ <!-- --> ] .pull-left[ <!-- --> ] --- class: duke-orange, middle ## Model 3: prestige ~ education + `log(income)` ### 1. Fit a model ### 2. Visualise the fitted model ### 3. Measure the strength of the fit ### 4. Residual analysis --- ## Model 3: prestige ~ education + `log(income)` ```r model3 <- lm(prestige ~ log(income) + education, data=train) summary(model3) ``` ``` Call: lm(formula = prestige ~ log(income) + education, data = train) Residuals: Min 1Q Median 3Q Max -17.3538 -4.7221 0.7046 4.3980 18.4684 Coefficients: Estimate Std. Error t value Pr(>|t|) (Intercept) -89.0742 13.2673 -6.714 2.62e-09 *** log(income) 10.4746 1.7069 6.136 3.16e-08 *** education 4.2117 0.3419 12.320 < 2e-16 *** --- Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1 Residual standard error: 7.296 on 79 degrees of freedom Multiple R-squared: 0.8238, Adjusted R-squared: 0.8194 F-statistic: 184.7 on 2 and 79 DF, p-value: < 2.2e-16 ``` --- ```r model3_fitresid <- augment(model3) model3_fitresid ``` ``` # A tibble: 82 x 9 .rownames prestige `log(income)` education .fitted .hat .sigma .cooksd <chr> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> 1 medical.techn… 67.5 8.55 12.8 54.4 0.0249 7.19 0.0283 2 welders 41.8 8.78 7.92 36.2 0.0337 7.31 0.00706 3 commercial.tr… 40.2 9.08 11.1 52.9 0.0202 7.20 0.0213 4 economists 62.2 8.99 14.4 65.9 0.0346 7.33 0.00326 5 farmers 44.1 8.20 6.84 25.6 0.0362 7.03 0.0834 6 receptionsts 38.7 7.97 11.0 40.9 0.0410 7.34 0.00139 7 sales.supervi… 41.5 8.92 9.84 45.8 0.0201 7.33 0.00243 8 mail.carriers 36.1 8.61 9.22 40.0 0.0164 7.33 0.00161 9 taxi.drivers 25.1 8.35 7.93 31.8 0.0245 7.30 0.00719 10 veterinarians 66.7 9.59 15.9 78.5 0.0636 7.21 0.0630 # … with 72 more rows, and 1 more variable: .std.resid <dbl> ``` --- If the variables used to fit the model are not included in data, then no .resid column will be included in the output. When you transform variables (say a log transformation), augment will not display .resid column. --- ### Add new variable (log.income) to the training set. ```r train$log.income. <- log(train$income) model3 <- lm(prestige ~ log.income. + education, data=train) model3_fitresid <- broom::augment(model3) model3_fitresid ``` ``` # A tibble: 82 x 10 .rownames prestige log.income. education .fitted .resid .hat .sigma .cooksd <chr> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> 1 medical.… 67.5 8.55 12.8 54.4 13.1 0.0249 7.19 0.0283 2 welders 41.8 8.78 7.92 36.2 5.59 0.0337 7.31 0.00706 3 commerci… 40.2 9.08 11.1 52.9 -12.7 0.0202 7.20 0.0213 4 economis… 62.2 8.99 14.4 65.9 -3.74 0.0346 7.33 0.00326 5 farmers 44.1 8.20 6.84 25.6 18.5 0.0362 7.03 0.0834 6 receptio… 38.7 7.97 11.0 40.9 -2.23 0.0410 7.34 0.00139 7 sales.su… 41.5 8.92 9.84 45.8 -4.31 0.0201 7.33 0.00243 8 mail.car… 36.1 8.61 9.22 40.0 -3.89 0.0164 7.33 0.00161 9 taxi.dri… 25.1 8.35 7.93 31.8 -6.67 0.0245 7.30 0.00719 10 veterina… 66.7 9.59 15.9 78.5 -11.8 0.0636 7.21 0.0630 # … with 72 more rows, and 1 more variable: .std.resid <dbl> ``` --- ### Plot of Residuals vs Fitted .pull-left[ Now - Model 3 <!-- --> ] .pull-right[ Before - Model 2 <!-- --> ] --- # Normality of residuals .pull-left[ ```r ggplot(model3_fitresid, aes(x=.resid))+ geom_histogram(colour="white")+ggtitle("Distribution of Residuals") ``` <!-- --> ] .pull-right[ ```r ggplot(model3_fitresid, aes(sample=.resid))+ stat_qq() + stat_qq_line()+labs(x="Theoretical Quantiles", y="Sample Quantiles") ``` <!-- --> ] --- ```r shapiro.test(model3_fitresid$.resid) ``` ``` Shapiro-Wilk normality test data: model3_fitresid$.resid W = 0.99414, p-value = 0.9747 ``` --- ## Plot of residuals vs predictor variables .pull-left[ <!-- --> ] .pull-left[ <!-- --> ] --- .pull-left[ ## Prestige vs income by type <!-- --> R code: ___________ ] .pull-right[ ## Prestige vs income by type <!-- --> R code: ______________ ] --- class: duke-orange, middle ## Model 4: prestige ~ education + log(income) + `type` ### 1. Fit a model ### 2. Visualise the fitted model ### 3. Measure the strength of the fit ### 4. Residual analysis --- ```r str(train) ``` ``` 'data.frame': 82 obs. of 8 variables: $ education : num 12.79 7.92 11.13 14.44 6.84 ... $ income : int 5180 6477 8780 8049 3643 2901 7482 5511 4224 14558 ... $ women : num 76.04 5.17 3.16 57.31 3.6 ... $ prestige : num 67.5 41.8 40.2 62.2 44.1 38.7 41.5 36.1 25.1 66.7 ... $ census : int 3156 8335 5133 2311 7112 4171 5130 4172 9173 3115 ... $ type : Factor w/ 3 levels "bc","prof","wc": 3 1 3 2 NA 3 3 3 1 2 ... $ id : int 31 79 51 14 67 42 50 43 99 25 ... $ log.income.: num 8.55 8.78 9.08 8.99 8.2 ... ``` --- ## Model 4: prestige ~ education + log(income) + `type` ```r model4 <- lm(prestige ~ log.income. + education + type, data=train) ``` --- ```r summary(model4) ``` ``` Call: lm(formula = prestige ~ log.income. + education + type, data = train) Residuals: Min 1Q Median 3Q Max -13.4996 -4.5672 0.5835 4.7882 18.1563 Coefficients: Estimate Std. Error t value Pr(>|t|) (Intercept) -75.5417 17.4258 -4.335 4.59e-05 *** log.income. 9.5470 2.2593 4.226 6.80e-05 *** education 3.5211 0.7523 4.681 1.29e-05 *** typeprof 6.6951 4.3625 1.535 0.129 typewc -1.8008 2.9820 -0.604 0.548 --- Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1 Residual standard error: 6.721 on 73 degrees of freedom (4 observations deleted due to missingness) Multiple R-squared: 0.8523, Adjusted R-squared: 0.8442 F-statistic: 105.3 on 4 and 73 DF, p-value: < 2.2e-16 ``` --- ## Model 4: prestige ~ education + log(income) + `type` ```r model4_fitresid <- augment(model4) head(model4_fitresid) ``` ``` # A tibble: 6 x 11 .rownames prestige log.income. education type .fitted .resid .hat .sigma <chr> <dbl> <dbl> <dbl> <fct> <dbl> <dbl> <dbl> <dbl> 1 medical.tec… 67.5 8.55 12.8 wc 49.3 18.2 0.0895 6.38 2 welders 41.8 8.78 7.92 bc 36.1 5.67 0.0373 6.73 3 commercial.… 40.2 9.08 11.1 wc 48.5 -8.34 0.0970 6.69 4 economists 62.2 8.99 14.4 prof 67.9 -5.66 0.0431 6.73 5 receptionsts 38.7 7.97 11.0 wc 37.6 1.05 0.0887 6.77 6 sales.super… 41.5 8.92 9.84 wc 42.5 -0.967 0.119 6.77 # … with 2 more variables: .cooksd <dbl>, .std.resid <dbl> ``` --- ## Plot of Residuals vs Fitted <!-- --> --- # Normality of residuals .pull-left[ ```r ggplot(model4_fitresid, aes(x=.resid))+ geom_histogram(colour="white")+ggtitle("Distribution of Residuals") ``` <!-- --> ] .pull-right[ ```r ggplot(model4_fitresid, aes(sample=.resid))+ stat_qq() + stat_qq_line()+labs(x="Theoretical Quantiles", y="Sample Quantiles") ``` <!-- --> ] --- ```r shapiro.test(model4_fitresid$.resid) ``` ``` Shapiro-Wilk normality test data: model4_fitresid$.resid W = 0.98621, p-value = 0.5658 ``` --- ## Plot of residuals vs predictor variables .pull-left[ <!-- --> ] .pull-left[ <!-- --> ] --- ## Multicollinearity ```r car::vif(model4) ``` ``` GVIF Df GVIF^(1/(2*Df)) log.income. 1.825493 1 1.351108 education 7.618522 1 2.760167 type 7.159248 2 1.635750 ``` VIFs larger than 10 imply series problems with multicollinearity. --- ## Detecting influential observations ```r library(lindia) gg_cooksd(model4) ``` <!-- --> --- # Influential outliers (cont.) ```r model4_fitresid %>% top_n(4, wt = .cooksd) ``` ``` # A tibble: 4 x 11 .rownames prestige log.income. education type .fitted .resid .hat .sigma <chr> <dbl> <dbl> <dbl> <fct> <dbl> <dbl> <dbl> <dbl> 1 medical.tec… 67.5 8.55 12.8 wc 49.3 18.2 0.0895 6.38 2 veterinaria… 66.7 9.59 15.9 prof 78.8 -12.1 0.0786 6.60 3 file.clerks 32.7 8.01 12.1 wc 41.7 -9.02 0.113 6.67 4 collectors 29.4 8.46 11.2 wc 42.9 -13.5 0.0591 6.57 # … with 2 more variables: .cooksd <dbl>, .std.resid <dbl> ``` --- ## Solutions - Remove influential observations, and re-fit model. - Transform explanatory variables to reduce influence. - Use weighted regression to down weight influence of extreme observations. --- # Hypothesis testing `\(Y = \beta_0 + \beta_1x_{1} + \beta_2x_{2} + \beta_3x_{3} + \beta_4x_{4} + \epsilon\)` ## What is the overall adequacy of the model? `\(H0: \beta_1= \beta_2 = \beta_3 = \beta_4 = 0\)` `\(H1: \beta_j \neq 0 \text{ for at least one } j, j=1, 2, 3, 4\)` ## Which specific regressors seem important? `\(H0: \beta_1= 0\)` `\(H1: \beta_1 \neq 0\)` --- # Making predictions Method 1 ```r head(test) ``` ``` education income women prestige census type id gov.administrators 13.11 12351 11.16 68.8 1113 prof 1 general.managers 12.26 25879 4.02 69.1 1130 prof 2 mining.engineers 14.64 11023 0.94 68.8 2153 prof 10 surveyors 12.39 5902 1.91 62.0 2161 prof 11 vocational.counsellors 15.22 9593 34.89 58.3 2391 prof 19 physicians 15.96 25308 10.56 87.2 3111 prof 24 ``` --- ```r test$log.income. <- log(test$income) predict(model4, test) ``` ``` gov.administrators general.managers mining.engineers 67.26199 71.33086 71.56331 surveyors vocational.counsellors physicians 57.67687 72.27901 84.14602 nursing.aides secretaries bookkeepers 35.60007 42.73588 42.49606 shipping.clerks telephone.operators sales.clerks 35.79196 36.60012 33.09309 service.station.attendant real.estate.salesmen policemen 33.60911 46.22149 49.75274 farm.workers textile.labourers machinists 25.50358 26.05783 39.56685 electronic.workers masons 34.34687 30.68619 ``` --- # Making predictions Method 2 ```r library(modelr) test <- test %>% add_predictions(model4) head(test, 4) ``` ``` education income women prestige census type id log.income. gov.administrators 13.11 12351 11.16 68.8 1113 prof 1 9.421492 general.managers 12.26 25879 4.02 69.1 1130 prof 2 10.161187 mining.engineers 14.64 11023 0.94 68.8 2153 prof 10 9.307739 surveyors 12.39 5902 1.91 62.0 2161 prof 11 8.683047 pred gov.administrators 67.26199 general.managers 71.33086 mining.engineers 71.56331 surveyors 57.67687 ``` --- # In-sample accuracy and out of sample accuracy ```r # test MSE test %>% add_predictions(model4) %>% summarise(MSE = mean((prestige - pred)^2, na.rm=TRUE)) ``` ``` MSE 1 41.17247 ``` ```r # training MSE train %>% add_predictions(model4) %>% summarise(MSE = mean((prestige - pred)^2, na.rm=TRUE)) ``` ``` MSE 1 42.2731 ``` --- ## Out of sample accuracy: model 1, model 2 and model 3 ```r # test MSE test %>% add_predictions(model1) %>% summarise(MSE = mean((prestige - pred)^2, na.rm=TRUE)) ``` ``` MSE 1 104.0138 ``` ```r # test MSE test %>% add_predictions(model2) %>% summarise(MSE = mean((prestige - pred)^2, na.rm=TRUE)) ``` ``` MSE 1 69.17519 ``` --- ```r # test MSE test %>% add_predictions(model3) %>% summarise(MSE = mean((prestige - pred)^2, na.rm=TRUE)) ``` ``` MSE 1 43.92756 ``` Model 4: 42.51 --- ## Modelling cycle .pull-left[ - EDA - Fit - Examine the residuals (multicollinearity/ Influential cases) - Transform/ Add/ Drop regressors if necessary ] .pull-right[ - Repeat above until you find a good model(s) - Use out-of-sample accuracy to select the final model ] --- ## Presenting results .pull-left[ - EDA - Final regression fit: - sample size - estimated coefficients and standard errors - `\(R_{adj}^2\)` - Visualizations (model fit, coefficients and CIs) ] .pull-right[ - Model adequacy checking results: Residual plots and interpretations - Hypothesis testing and interpretations - ANOVA, etc. - Out-of sample accuracy ] - Some flexibility is possible in the presentation of results and you may want to adjust the rules above to emphasize the point you are trying to make. --- # Other models - Decision trees - Random forests - XGBoost - Deep learning approaches and many more.. --- class: center, middle Slides available at: hellor.netlify.app All rights reserved by [Thiyanga S. Talagala](https://thiyanga.netlify.com/)