For this project I will be using a data set from Kaggle. The data is of the top songs streamed on Spotify during the year 2023. There are 943 unique entries that will be a part of this analysis. The goal of this project is to see if there is any correlation between the release date of a song and the number of streams it receives.

##Cleaning Data
dataset = read.csv("spotify-2023.csv")
str(dataset)
summary(dataset)

dataset = na.omit(dataset)##omit rows with missing values

dataset = dataset[!duplicated(dataset), ]
##ignore duplicates
dataset$streams <- gsub("[^0-9]", "", dataset$streams)
dataset$streams = as.numeric(dataset$streams)
##convert streams to numeric
dataset <- dataset[dataset$streams != max(dataset$streams), ]

We begin by loading and cleaning the data to ignore duplicated and missing values. We also convert the “Streams” variable to numeric for later.

summary(dataset$released_year)
summary(dataset$released_month)
summary(dataset$released_day)

We follow up by taking the summary of the different data we will be using in this analysis. We can see that although it is streaming data from 2023, the mean year of song is 2018. As far as months go, the mean is just over 6 which is right in the middle. For days, the mean is just under 14, so there are slightly more songs in the data release near the start of the month. \[\text{Mean of Year} = 2018 \\ \text{Mean of Month} = 6.034 \\ \text{Mean of Day} = 13.93\]

cor_matrix <- cor(dataset[, c("released_year",
                              "released_month",
                              "released_day",
                              "streams")])
print(cor_matrix)

##                released_year released_month released_day     streams
## released_year     1.00000000     0.07105497   0.16973315 -0.23080298
## released_month    0.07105497     1.00000000   0.07839121 -0.02493793
## released_day      0.16973315     0.07839121   1.00000000  0.01059794
## streams          -0.23080298    -0.02493793   0.01059794  1.00000000

We can see that our correlation between streams, and our other variables are mostly negative. The correlation between streams and released year indicates that older songs tend to have more streams than newer songs. The correlation between streams and released month indicates that songs released earlier in the year tend to have more streams than songs released later in the year. And being the only positive, the correlation between streams and released day indicates that songs released earlier in the month have less streams compared to those released later in the month.

dataset$released_year <- as.numeric(as.character(dataset$released_year))
lm_model <- lm(streams ~ released_year +
                 released_month + released_day, data = dataset)

## 
## Call:
## lm(formula = streams ~ released_year + released_month + released_day, 
##     data = dataset)
## 
## Residuals:
##        Min         1Q     Median         3Q        Max 
## -1.477e+09 -3.388e+08 -1.981e+08  1.540e+08  3.160e+09 
## 
## Coefficients:
##                  Estimate Std. Error t value Pr(>|t|)    
## (Intercept)     2.529e+10  3.327e+09   7.602 6.98e-14 ***
## released_year  -1.229e+07  1.651e+06  -7.444 2.19e-13 ***
## released_month -1.917e+06  5.043e+06  -0.380    0.704    
## released_day    3.209e+06  1.978e+06   1.622    0.105    
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## Residual standard error: 551600000 on 948 degrees of freedom
## Multiple R-squared:  0.05596,    Adjusted R-squared:  0.05298 
## F-statistic: 18.73 on 3 and 948 DF,  p-value: 8.252e-12

This regression analysis shows us that for every year that advances, a song is likely to have 12,290,000 less streams than the previous year given other factors are constant. Our released year also has a p value where \[ p < 0.05 \] So we can say that the released year and streams relationship is statistically significant. However we can’t say the same for released month and released day who both have p values where \[ p > 0.05 \]

lm_model_year <- lm(streams ~ released_year, data = dataset)
summary(lm_model_year)

Because the multiple r-squared value was so low, and the p values for release month and release day were so high. I ran regression with just release_year. The P value comes in under 0.5 and the R-Squared value comes in at 0.4837. Showing that there is some significant relationship between streams and release year.

The regression results show us that the release year has a statistically significant and negative impact on streams. Release month and day however, do not have a significant impact on streams. The multiple R-Squared value is 0.05596, which indicated that around 5.6% of the variance is due to release year, release month, and release day combined. However, when I did regression analysis with just release year, the R-Squared value came back as 0.4837. This shows us that 48.37% of the variance in streams is explained by just release year.