This is a random data set that I pull from Kaggle. It contained a list of the top trending videos on YouTube from the US in a random day back in 2017. I want to build a linear regression model to see if number of likes can be predicted by views.
library(tidyverse)
## Warning: package 'tidyverse' was built under R version 3.6.3
## -- Attaching packages --------------------------------------- tidyverse 1.3.0 --
## v ggplot2 3.3.3 v purrr 0.3.3
## v tibble 3.1.0 v dplyr 1.0.5
## v tidyr 1.1.3 v stringr 1.4.0
## v readr 1.3.1 v forcats 0.4.0
## Warning: package 'ggplot2' was built under R version 3.6.3
## Warning: package 'tibble' was built under R version 3.6.3
## Warning: package 'tidyr' was built under R version 3.6.3
## Warning: package 'readr' was built under R version 3.6.2
## Warning: package 'purrr' was built under R version 3.6.2
## Warning: package 'dplyr' was built under R version 3.6.3
## Warning: package 'stringr' was built under R version 3.6.2
## Warning: package 'forcats' was built under R version 3.6.2
## -- Conflicts ------------------------------------------ tidyverse_conflicts() --
## x dplyr::filter() masks stats::filter()
## x dplyr::lag() masks stats::lag()
df = read.csv('USvideos.csv', stringsAsFactors = FALSE)
head(df)
## video_id title
## 1 2kyS6SvSYSE WE WANT TO TALK ABOUT OUR MARRIAGE
## 2 1ZAPwfrtAFY The Trump Presidency: Last Week Tonight with John Oliver (HBO)
## 3 5qpjK5DgCt4 Racist Superman | Rudy Mancuso, King Bach & Lele Pons
## 4 puqaWrEC7tY Nickelback Lyrics: Real or Fake?
## 5 d380meD0W0M I Dare You: GOING BALD!?
## 6 gHZ1Qz0KiKM 2 Weeks with iPhone X
## channel_title category_id views likes dislikes comment_count
## 1 CaseyNeistat 22 748374 57527 2966 15954
## 2 LastWeekTonight 24 2418783 97185 6146 12703
## 3 Rudy Mancuso 23 3191434 146033 5339 8181
## 4 Good Mythical Morning 24 343168 10172 666 2146
## 5 nigahiga 24 2095731 132235 1989 17518
## 6 iJustine 28 119180 9763 511 1434
dim(df)
## [1] 40949 8
cor(df %>% dplyr::select(views, likes, dislikes, comment_count)) %>% round(., 3)
## views likes dislikes comment_count
## views 1.000 0.849 0.472 0.618
## likes 0.849 1.000 0.447 0.803
## dislikes 0.472 0.447 1.000 0.700
## comment_count 0.618 0.803 0.700 1.000
chart1 = with(df,
plot(views, likes,
xlab = 'views',
ylab = 'likes',
main = 'Views x Likes'))
chart1
## NULL
mod = lm(likes ~ views, data = df)
summary(mod)
##
## Call:
## lm(formula = likes ~ views, data = df)
##
## Residuals:
## Min 1Q Median 3Q Max
## -2249306 -17006 -11581 675 3019010
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 1.221e+04 6.271e+02 19.47 <2e-16 ***
## views 2.629e-02 8.079e-05 325.38 <2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 120900 on 40947 degrees of freedom
## Multiple R-squared: 0.7211, Adjusted R-squared: 0.7211
## F-statistic: 1.059e+05 on 1 and 40947 DF, p-value: < 2.2e-16
par(mfrow = c(2, 2))
plot(mod)
The residual plot and the q-q plot definitely suggests that we should not build a linear regression model using the variables without any sort of transformation. Perhaps we should first bucket the variables, e.g. cut views into deciles and then build a model on each. Or, remove any outlier and then apply Box-Cox to transform the data to make sure it becomes constant variance and suitable for linear regression.