Report Writing In Logistic Regression

Report Writing In Logistic Regression

Title

Introduction

Set up the problem.

What are you trying to predict or model, and for what purpose ? You use logistic regression to find the probability of a success given some explanatory

variables. One possible question that you can answer using logistic regression, does

success increase in SAT test if I increase my study time? If we increase the dosage of a

medication will more animals get better? Will likelihood of success increase as the number of hours of study increase?

Method and Results

Pre-model design

o Each variable in the data set has to be described, and a summary statistics has to be

performed. Size of your sample, general description of the dependent and independent

variables: numeric, ordinal, nominal.

o Dependency between the response and the explanatory variables has to be established,

when appropriate. Yes, the response variable has only two possible values, 0 and 1.

o If the explanatory variable is ordinal or nominal you will need to perform chi-square test for

independence, and also plot a bar chart, or mosaic.

o If the explanatory variable is numerical, at this moment, you have nothing to do.

o If you have 2 or more explanatory variables numerical you have to plot them against each

other to check for co-linearity.

Model Design

Given what you discovered in the pre-model phase, get the logistic regression equation The model is good if:

Step 1:

Take difference between Null and Residual deviance. dnull_residual (1)

Take difference between dfs of the Null and Residual deviance. ddf_null_residual (2)

If dnull_residual <= ddf_null_residual then calculate pchisq(dnull_residual, ddf_null_residual )

Else 1 - pchisq(dnull_residual, ddf_null_residual )

Note: This how you do it in R: with(mylogit, pchisq(null.deviance - deviance, df.null - df.residual,

lower.tail = FALSE))

If value pess than 0.05 than the model performed better than the empty model

上一页第2页下一页