0:03 When you go on a quest and that quest is really awesome, it's a StatQuest. Yeah, yeah, yeah. Hello and welcome to StatQuest. StatQuest is brought to you by the friendly folks in the genetics department at the University of North Carolina at Chapel Hill. Today we're going to talk about fitting a line to data, aka leased squares, aka Linear regression. Now let's get to it. Okay, you worked really hard, you did the experiment, and now you got some data. Here it is, plotted on an x-y graph. We usually like to add a line to our data so we can see what the trend is. But is this the best line we should use?
0:53 Or does this new line fit the data even better? Or what about this line? Is it better or worse than the other options? A horizontal line that cuts through the average y value of our data is probably the worst fit of all. However, it gives us a good starting point for talking about how to find the optimal line to fit our data. So now let's focus on this horizontal line. It cuts through the average y value, which is 3.5. Let's just call this point B, because different data sets will have different average values on the y-axis. That is to say, the y value for this line
1:39 is B, and for this particular data set, B equals 3.5. We can measure how well this line fits the data by seeing how close it is to the data points. We'll start with the point in the lower left-hand corner of the graph with coordinates x1, y1. We can now draw a line from this point up to the line that cuts across the average y value for this data set. The distance between the line and the first data point equals b minus y1. The distance between the line and the second data point is b minus y2. So far,
2:26 the total distance between the data points and the line is the sum of the two distances. And we can calculate the distance between the line and the third point. That equals b minus y3. Now we've added the third distance to our total sum. The distance for the fourth point is b minus y4. Note, y4 is greater than b because it's above the horizontal line, so this value will be negative. That's no good, since it will subtract from the total and make the overall fit appear better than it really is. The fifth data point is even higher relative to the
3:12 horizontal line. This distance is going to be very negative. Back in the day, when they were first working this out, They probably tried taking the absolute value of everything and then discovered that it made the math pretty tricky. So they ended up squaring each term. Squaring ensures that each term is positive. Here's the equation that shows the total distance the data points have from the horizontal line. In this specific example, 24.62 is our measure of how well this line fits the data. It's called the sum of squared residuals because the residuals are the differences between the real data and
3:59 the line, and we are summing the square of these values. Now let's see how good the fit is if we rotate the line a little bit. In this case, the sum of squared residuals equals 18.72. This is better than before. Does this fit improve if we rotate a little more? Yes! The sum of squared residuals now equals 14.05. That value keeps going down the more we rotate the line. What if we rotate the line a whole lot? Well, as you can see, the fit gets worse. In this case, the sum of squared residuals is 31.71.
4:48 So there's a sweet spot in between horizontal and two vertical. To find that sweet spot, let's start with the generic line equation. This is y equals ax or a times x plus b. a is the slope of the line and b is the y-intercept of the line. That's the location on the y-axis that the line crosses when x equals 0. We want to find the optimal values for A and B so that we minimize the sum of squared residuals. In more general math terms, the sum of squared residuals is this complicated mathematical equation. But it's
5:34 actually not that complicated. This first part is the value of the line at x1. And this second part is the observed value at x1. So really all we're doing in this part of the equation is calculating the distance between the line and the observed value. So this is no big deal. Since we want the line that will give us the smallest sum of squares, this method for finding the best values for a and b is called least squares. If we plotted the sum of squared residuals versus each rotation, we'd get something like this, where on the y-axis we have the sum of squared residuals, and on the x-axis we've got each
6:21 different rotation of the line. We see that the sum of squared residuals goes down when we start rotating the line, but that it's possible to rotate the line too far and the sum of squared residuals starts going back up again. How do we find the optimal rotation for the line? Well, we take the derivative of this function. The derivative tells us the slope of the function at every point. The slope at the point on the far left side is pretty steep. As we move to the right, we see that the slope isn't as steep. The slope at the best point, where we have the least squares, is zero.
7:06 After that, the slope starts getting steep again. Let's go back to that middle point where we have the least squares value and the slope is zero. Remember, the different rotations are just different values for A, the slope, and B, the intercept. We can use a 3D graph to show how different values for the slope and intercept result in different sums of squares. In this graph, The intercept is the z-axis, so it's going back sort of deep into your computer screen. And if we select one value for the intercept, for example, assume we set the intercept value to be 3, then we could change values
7:53 for the slope and see how an intercept of 3 plus different values for the slope would affect the sum of squared residuals. Anyways, we do that for bunches of different intercepts and slopes. Taking the derivatives of both the slope and the intercepts tells us where the optimal values are for the best fit. No one ever solves this problem by hand. This is done on a computer. So for most people, it's not essential to know how to take these derivatives. However, it's essential to understand the concepts. Big important concept number one. We want to minimize the square of the distance
8:39 between the observed values and the line. Big important concept number two. We do this by taking the derivative and finding where it is equal to zero. The final line minimizes the sums of squares. It gives the least squares between it and the real data. In this case, The line is defined by the following equation. Y equals 0.77 times X plus 0.66. Hooray! We've made it to the end of another stat quest. Tune in next time for another exciting adventure in statistics land.
