One of the primary goals of analyzing sports data is to understand the relationship between two factors. For example, we may be interested in individual play performance and team performance. When individual players' performance improve, would a team overall performance improve as well? We may also be interested in the viewership in a team performance. When the team performs better, would viewership go up as well? Similarly, we may be interested in the ticket price and attendance. When ticket price increases, would attendance increase or decrease? To view the relationship between two variables, we can graph a scatter plot of the two variables in Python. For example, it may be reasonable to suspect that there may be some relationship between a number of assists and a number of field goals made in basketball. We can create a scatter plot using the plot dot scatter function in Python. I specify the number of assists and the variable AST in the x-axis and the number of field goals made the variable FGM in the y-axis. From this scatter plot, we can see that as the number of assists increase, it appears that the number of field goals may also increase as well. To better visualize this relationship, we can also add a regression line to the graph. A regression line will fit the data point into a linear line. In this case, the number of field goals made will be a linear function of the number of assist. We'll go full regression analysis in more detail in the upcoming weeks. To add a regression line, we need to first introduce a new library, called Seaborn. Seaborn is a data visualization library in Python that provides a high level interface for drawing informative statistical graphs. We import Seaborn as sns. We can use the regplot function, R-E-G-P-L-O-T function in Seaborn Library to graph a regression line as well as the scatter plot. In this line of code, we tell Python to put the number of assist, AST, in the x-axis and the number of field goals made, FGM, on the y-axis. We also specify the data, which is the NBA games. We also need to specify the marker. Similarly, we can use the PY plot function from the map-plot library to add labels and titles into our graph. Now this graph with the regression line gives us a more clear picture that as the number of assist increase, the number of field goals made also increase. In this case, we can say that there is some positive relationship between the two variables. We can quantify the linear relationship using correlation analysis. In particular, we can use a correlation coefficient, which is a single number, to summarize this linear relationship. To introduce correlation coefficient, we need to first introduce covariance, which measures the joint variability of two variables. Covariance is defined as the average of the product of the deviation form the individual means. The sign of the covariance shows a tendency in the linear relationship between the two variables whether the two variables are moving the same or opposite direction. When the covariance equals to zero, we say that the two variables are not linearly related. This could mean that when x changes, there is no change in y. This could also means that when x changes, y is changing but in a non-linear fashion. When the covariance is positive, we say that the two variables are positively related. This means that when x increases, y increases as well. Similarly, when the covariance is negative, it means that the two variables are negatively related. When x increases, y would decrease. It is sometimes hard to interpret covariance, in particular, the size of the covariance. For example, if we have a covariance equals to 100, equals to 10, or equals to 0.5, can we say that the correlation between the two variables is strong or weak? One problem with covariance is that the covariance depends on the unit of measurement. Therefore, it is hard to interpret the size of the covariance. Instead, we typically only interpret the sign of the covariance, whether it's positive or negative. Correlation coefficient, on the other hand, is a standardized measure. Correlation coefficient is defined as the covariance divided by the product of the standard deviations of the two variables. Correlation coefficient, similar to covariance measures the linear relationship. Correlation coefficient does not depend on the units of measurement, and it's always between negative one and one. If the correlation coefficient is greater than zero, we say that the two variables are positively related. That means that as the value of one variable increases, the value of the other variable increase as well. When the correlation coefficient is negative, that means that the two variables are inversely related. When the value of one variable increases, the value of the outer variable would decrease. When the correlation coefficient equal to zero, we say that there is no linear relationship between the two variables. In some rare cases, when the correlation coefficient equals to one or when the correlation coefficient equals to negative one, then there would be perfect linear relationship between the two variables. Again, it's important to emphasize that correlation coefficient only measures linear relationship when the absolute value of the correlation coefficient goes to zero, it does not necessarily mean that there's no relationship between the two variables. It just means that these two variables are not linearly related. In correlation coefficient only measures whether the two variables are moving in the same or opposite direction. It does not say anything about causal relationship between the two variables. In Python, we can use the corr function, corr to calculate correlation coefficient. We specify the first variable, NBA games, number of assists, ast, and use the corr function and inside a corr function we specify the other variable, NBA games are fgm. The correlation coefficient between the number of assists and the number of field goal made is about 0.7. We can say that there is a positive correlation between the two variables. Let's also investigate the relationship between the number of assists and the number of field goals attempted. We first graph a scatter plot with a regression line as well to visualize the relationship between the two variables, ast and in this time, fga, field goal attempted. As we can see from this graph, as the number of assists increase, there is some slight increase in the field goal attempted. Especially when we compare this graph between the number of assists and field goal attempted, to the graph between the number of assists and field goals made. We can see that the slope in the second graph, it's much smaller than the previous graph. Now let's also calculate the correlation coefficient between the number of assists and field goals attempted. This correlation coefficient, it's about 0.36, which is much smaller than the correlation coefficient between the number of assists and field goals made, which is 0.7. Based on the graph and the correlation coefficient, we can see that there's only a very small positive relationship between the number of assists and field goals attempted. We can also grab the scatter plot by group using the hue option, hue option. Let's say we want to separate by the result of the game, whether the team won the game or lost the game. We'll continue to produce a scatter plot with regression line between the number of assists and number of goals attempted. In this case, we'll use lm plot function instead of red plot function in the seaborn library. The lm plot function combines red plot function and Facet Grid. Facet Grid produces multi-plot grid for plotting conditional relationships. Therefore Facet Grid allows us to separate the dataset into multiple panels based on specified conditions to regionalize the relationship between multiple variables. Inside this lm plot function, we will add an option. Ek hue equals to wl. As we can see from this graph, there's no significant difference in the relationship between the number of assists and field goals attempted when the team won the game and when a team lost the game. Lastly, we can also construct a correlation coefficient table for all the numerical variables. We can simply specify the data frame name, NBA games and use the corr function and the correlation coefficient we introduce is a Pearson correlation coefficient. We need to specify method equals to Pearson. This table shows the correlation coefficient between any two numerical variables in our dataset.
