2181S Introduction to Data Mining Lecture 5 Scatter Plots | Scatter Plots Scatter plot (or scatter diagram) shows the position of all of the cases in an x-y coordinate system. The relationship between two interval variables can be identified from Scatter graph. — The independent variable is plotted on the x-axis, or the horizontal axis. — The dependent variable is plotted on the y-axis, or the vertical axis A dot in the body of the chart represents the intersection of the data on the x and y-axis Use of Scatter Plots + The purpose of Scatter Plot: — Perform regression analysis and Correlation checks + With the information obtained from the above techniques, can answer the following questions that identify the relationships between variables: — Is there a relationship between the two variables? (The two variables usually have range of values with scale or ordinal measure) — How strong is the relationship? — What is the direction of the relationship? Finding Out the Relationship — + The overall pattern of data points summarizes the nature of the relationship between the two variables. + The clarity of the pattem formed by the data points can be enhanced by drawing a straight line through the cluster such that the line touches every dot or comes as close to doing so as possible. + This summarizing line is called “regression line.” — It finds the line that best predicts Y from values of XSample Scatter Plot Direction of Relationship The pattern of the points on the scatter plot + Positive vs Negative Relations shows the relationship between the variables. — The dependent variable increases when the The regression line independent variable increases > Positive relation makes it easier — The dependent variable decreases when the to understand the independent variable increases > Negative relation relationship. ip i s AS i y i ry | i i : | A » a : Fostive Relation] [M@awvercawo] =, Evidence of Relationship | Strength of Relationship The angle between the regression line and the > ‘The relationship between the two variables can horizontal x-axis provides evidence of a be drawn from the scatter plot relationship. If there is no relationship, the — A horizonial or vertical line means no relation | regression line will be parallel to the x or y-axis. — Avery flat or steep line usually means slight relation [> Perfect Relator : i [eon eee ET However, by looking at the angle of the regression line, you can be easily misled by the scaling of the graph! — Next slide shows an exampleEffect of Scaling on Scatter Plot Slope/Intercept of Regression Line | + The scale of the x and y-axis can affect the view | + The value of the slope b for the regression line | of the scatter plot! - can be found by the formula: | | 7 La- Ladin | | it ee = : 2 Satay ln | | spe |= where (3, y) is the value of data point i, and nis the | [ivaetaitiwepbtearefem) = | number sf data points aay | | aes cece meren) os + The y-intercept a can be found by: LPL | | | should be the same. Sealing : 9 . | eangveyouthewongvewl | "ss yp ww r Equation of the Regression Line | Important Note | | + Amore objective measure about the relation + If we interchange the X and Y data, the scatter | | between the two variables is to observe the plot will be transposed. However, the regression | equation of the regression line line is notan exact transpose. — We know the line can be expressed as y = a + bx ~ The transpose of y = 1.6x + 1 is y = 0.625x - 0.625, |= aand b are constants, where b is the slope and a is but the ragression lin is y = 0.597x ~ 0.4627 | the y-intercept (value of y is a when x is 0) — The slope gives a more objective measure of the | steepness of the graph | — Positive relation has positive slope, negative relation has negative slopeWhat is R2? + You may notice R? appears in Excel and SPSS — Itis a measure of how well the values are fitted to the regression line (goodness of fit) — A value closer to 1 means the data points are closer to the regression line. —Avalue 1 means all the data points are on the regression line. + In fact, Ris the square of the correlation coefficient ~ R has the value between O and 1 Correlation Coefficient + The (Pearson) correlation coefficient (also written as r) has been talked about in Lecture 4: ~ Correlation(x, y): om a2, See a0i-m) ee vihere iB -H, 2 nol (3 ar tha sev of and ¥ resp. Correlation coefficient has a value between -1 and +4 Meaning of Correlation Coefficient Correlation Coefficient Meaning No correlation Slight correlation Low/Small correlation ‘Moderate to Substantial 44-070 correlation 0.71-0.90 Strong/High correlation 0.91-0.99 Near Perfect correlation 1.00 Perfect correlation + Similar for negative values What do all these things mean? + Correlation coefficient rand R? give the relation between 2 variables — ralso shows the direction of relationship — R? closer to 1 > more relation between 2 variables + The regression line is the “best fit’ line of the points on the scatter plot — We would also like to know the average error of each point with respect to the regression lineHow to find the Error? + The y ertor for every data point is: — For every data point, get the x-coordinate. Find the difference of the expected y value with the actual y value ~ Example: For the data expected actual y=2, Summing the Errors + Note that the summation of y errors for every data point in the scatter plot is always 0 Standard Error of Regression Line | + The Standard Error S, y of the regression line is defined by the formula: 7 yom Saf n-2 In the previous example, the standard error of the regression line is sqrt(1.20/3) = 0.632 Question: What is the standard error for the transpose of the scatter plot (i.e., interchanging X and Y)? Transpose of the Scatter Plot | + The transpose of the scatter plot has regression | line with equation y = 0.697% — 0.4627 | x_[Expected y (y)] Actual y | Error | Error? i 7 ‘ 1 26 20 06! 0.36 2 1 : cage 2 42 5.0 08] 0.64 5 2 ‘ 3 58 60 -0.2| 0.04 6 3 2 [sf 7a 70} 04] 076 Tope : 5 a0 | 90 0.0/ 0.00 9 | 5 ‘ 5 Total oo] 1.20Standard Error x_[Expectedy (y)| Actualy | Eror | Erro® 2 [07313 1.0 _|-0.2687| 5 2.5223 2.0 | 0.5223] 0.2728| 6 | 31193 3.0 | 0.1193) 0.0142 7 | 3.7163 4.0 |-0.2837| 0.0805 9 | 4.9103 5.0 _|-0.0897| 0.0080 Total 0.0] 0.447 + Standard error is sqrt(0.4477/3) = 0.3863