Fitting curves through points: Regression - PowerPoint PPT Presentation

1 / 24
About This Presentation
Title:

Fitting curves through points: Regression

Description:

... of New Accounts versus Number of Sales Staff during a particular one year period ... Forensic Example (FSI, 1994) Stat 486. 14. Example: Growth.jmp. Exclude ... – PowerPoint PPT presentation

Number of Views:98
Avg rating:3.0/5.0
Slides: 25
Provided by: jamese170
Category:

less

Transcript and Presenter's Notes

Title: Fitting curves through points: Regression


1
Chapter 8
  • Fitting curves through points Regression

2
Introduction
  • This chapter considers data comprising a
    continuous independent variable x and a
    continuous response variable y
  • Simplest model is a straight line
  • Use the model to test the significance of the
    linear relationship and make predictions
  • Least squares most common way to choose the best
    line
  • demoLeastSquares.jsl

3
(No Transcript)
4
Example
  • Data from 232 branches of an East Coast bank
  • First look at Number of New Accounts versus
    Number of Sales Staff during a particular one
    year period

5
(No Transcript)
6
  • Similar to ANOVA except the fitted means are
    constrained to fall on a line

7
Linear Regression
ANOVA
8
Regression Model
  • Number of new accounts
  • 179.4 (78.9 x sales staff) residual
  • Parameter estimates subject to sampling
    variability
  • Can test hypotheses, compute confidence
    intervals, etc.

9
Confidence Curves Fit
10
Regression Model Assumptions
  • At each level of x, have a random sample of ys
  • At each level of x, the ys are normally
    distributed
  • All those normal distributions have the same
    variance

11
Residual Plots
  • Use residual plots to check assumptions
  • Residuals should be approximately normally
    distributed
  • Residuals when plotted against x or y should show
    no particular pattern

12
(No Transcript)
13
Forensic Example (FSI, 1994)
14
Example Growth.jmp
Exclude points to the left???
15
Polynomial Models
16
Transformations
fit special Splines? Piecewise?
17
Tukeys Bulging Rule
x2 y2
y2, sqrt x log x, 1/x
x2, sqrt y log y, 1/y
sqrt x, sqrt y, log x, log y, 1/x, 1/y
18
Another example Cell Phone Use
Which transformation? polynomial?
19
Always look at the data!
20
(No Transcript)
21
Prediction Cottages Example
22
Prediction for 3,000 sq. feet?
  • Two components of uncertainty
  • Dont know where the true line is
  • Even if we know where the true line is, the data
    are normally distributed about this line
  • Confidence curves fit only show uncertainty
    about the true line
  • Confidence curves indiv include both
    components of uncertainty

23
Fit
Indiv
24
Revisit Cell Phone Use
Prediction for Dec, 1999 (period 31)?
Write a Comment
User Comments (0)
About PowerShow.com