Title | Week 3 - Summary of lecture 3 |
---|---|
Author | Josh Garvey |
Course | Introduction to Bus Analytics |
Institution | University College Dublin |
Pages | 3 |
File Size | 111.9 KB |
File Type | |
Total Downloads | 30 |
Total Views | 145 |
Summary of lecture 3...
Week 3 Correlation detects a linear relationship if any between two variables Linear regression finds the line or plane which expresses this relationship Y = a + bx + error The goodness of fit of a regression revolves around the error term Use the independent variables to model the dependent variables Multiple regression Dependent variable - y (eg. Price) Independent variable - x1 / x2 / x3 etc (eg. Beds / area / baths) Model: y = a + b1x1 + b2x2 + b3x3 + error Use =linest to find the intercept (a) and b1 / b2 / b3 The first value calculated is the final value in the model Sample Model Independent variable = sales Dependent variables = area / distance y = -8.24 + (0.018)(area) + (0.137)(distance) When area and distance are zero the expected sales are -8.24 If area increases by one square foot the sales increase by 0.018K provided distance remains constant For every increase of one foot away from a competitor sales increase by 0.139K provided area remains constant Using linest on excel the first result is the last variable R = correlation between the prediction and the actual values R^2 = what proportion of the variability of y is caused by the other variables Multiple regression takes into account a number of independent variables. It is presented as a plane when graphed The process for finding the model is the same regardless of the number f variables 1. Plot scatterplots of each variable against y to establish any relationship 2. Fit models 3. Examine plots of the error terms against y and against the independent variables to see if any patterns emerge For exponential models you can take the log of y or x to create a linear model Log(y) for slide 23
Use linear regression to find a and b Y1 = loge(y) our estimate Y(hat) = eY(hat)1
Log(x) for slide 26
Use linear regression to find a and b X1 = loge(x) our estimate x = eX1
If the model has a quadratic relationship, take the square of y (Slide 27) How to transform data sets to enhance linearity Conduct standard regression on the raw data Construct an error plot If the pattern is random don’t transform the data If there are patterns transform the data Calculate the coefficient of determination (R2) using (y, y(hat)) Transform either the dependent, independent or both variables Conduct regression analysis Calculate R2 on transformed variables
Logistic Equation
1 ( −x ) 1+ e...