stata statistics

Regression in Stata

Ista Zahn

Harvard MIT Data Center

January 24 2013

�e Institutefor Quantitative Social Scienceat Harvard University

Ista Zahn (Harvard MIT Data Center) Regression in Stata January 24 2013 1 / 35

Outline

1 Introduction

2 Univariate regression

3 Multiple regression

4 Interactions

5 Exporting and saving results

6 Wrap-up


Introduction

Topic

1 Introduction



4 Interactions


6 Wrap-up


Introduction

Organization

Please feel free to ask questions at any point if they are relevant tothe current topic (or if you are lost!)There will be a Q&A after class for more specific, personalizedquestionsCollaboration with your neighbors is encouragedIf you are using a laptop, you will need to adjust paths accordinglyMake comments in your Do-file rather than on hand-outsSave on flash drive or email to yourself


Introduction

Copy the workshop materials to your home directory

Log in to an Athena workstation using your Athena user name andpasswordClick on the “Ubuntu” button on the upper-left and type “term” asshown below

Click on the “Terminal” icon as shown aboveIn the terminal, type this line exactly as shown:

cd; wget http://tinyurl.com/stata-stats-zip; unzip stata-stats-zip

If you see “ERROR 404: Not Found”, then you mistyped the command– try again, making sure to type the command exactly as shown


Introduction

Launch Stata on Athena

To start Stata type these commands in the terminal:

add stataxstata

Open up today’s Stata scriptIn Stata, go to Window => New do file => OpenLocate and open the StatStatistics.do script in the StataStatisticsfolder in your home directory

I encourage you to add your own notes to this file!


Introduction

Today’s Dataset

We have data on a variety of variables for all 50 statesPopulation, density, energy use, voting tendencies, graduation rates,income, etc.We’re going to be predicting SAT scoresUnivariate Regression: SAT scores and Education ExpendituresDoes the amount of money spent on education affect the mean SATscore in a state?Dependent variable: csatIndependent variable: expense


Introduction

Opening Files in Stata

Look at bottom left hand corner of Stata screen – This is the directoryStata is currently reading fromFiles are located in the StataDatMan folderStart by changing directory and loading the data

// change directorycd "C:/Users/dataclass/Desktop/StataStatistics"// use dir to see what is in the directory:dir// tell Stata to use the states data setuse states.dta


Introduction

Steps for Running Regression

1 Examine descriptive statistics2 Look at relationship graphically and test correlation(s)3 Run and interpret regression4 Test regression assumptions


Univariate regression

Topic

1 Introduction



4 Interactions


6 Wrap-up



Univariate Regression: Preliminaries

We want to predict csat scores from expenseFirst, let’s look at some descriptives

// generate summary statistics for csat and expensesum csat expense// look at codebokcodebook csat expense



Univariate Regression

Look at scatterplots, compute correlation matrix, and regress SATscores on expenditures

// graph expense by csattwoway scatter expense csat

// correlate csat and expensepwcorr csat expense, star(.05)

// run the regressionregress csat expense



Linear Regression Assumptions

Assumption 1: Normal DistributionThe errors of regression equation are normally distributedAssumption 2: Homoscedasticity (The variance around the regressionline is the same for all values of the predictor variable)Assumption 3: Errors are independentAssumption 4: Relationships are linear



Homoscedasticity



Testing Assumptions: Normality

A simple histogram of the residuals can be informative

// graph the residual values of csatpredict resid, residualhistogram resid, normal



Testing Assumptions: Homoscedasticity

rvfplot


Multiple regression

Topic

1 Introduction



4 Interactions


6 Wrap-up


Multiple regression

Multiple Regression

Just keep adding predictorsLet’s try adding some predictors to the model of SAT scoresincome :: % students taking SATspercent :: % adults with HS diploma (high)


Multiple regression

Multiple Regression Preliminaries

As before, start with descriptive statistics and correlations

// descriptive statisticssum income percent high

// generate correlation matrixpwcorr csat expense income percent high

// regress csat on exense, income, percent, and high\regress csat expense income percent high


Multiple regression

Exercise 1: Multiple Regression

Open the datafile, states.dta.1 Select a few variables to use in a multiple regression of your own.

Before running the regression, examine descriptive of the variables andgenerate a few scatterplots.

2 Run your regression3 Examine the plausibility of the assumptions of normality and

homogeneity


Interactions

Topic

1 Introduction



4 Interactions


6 Wrap-up


Interactions

Interactions

What if we wanted to test an interaction between percent & high?Option 1: generate product terms by hand

// generate product of percent and highgen percenthigh = percent*highregress csat expense income percent high percenthigh


Interactions

Interactions

What if we wanted to test an interaction between percent & high?Option 2: Let Stata do your dirty work

// use the # sign to represent interactionsregress csat percent high c.percent#c.high// same as . regress csat c.percent##high


Interactions

Categorical Predictors

For categorical variables, we first need to dummy codeUse region as example

Option 1: create dummy codes before fitting regression model

// create region dummy codes using tabtab region, gen(region) // could also use gen / replace

//regress csat on regionregress csat region1 region2 region3


Interactions

Categorical Predictors

For categorical variables, we first need to dummy codeUse region as example

Option 2: Let Stata do it for you

// regress csat on region using fvvarlist syntax// see help fvvarlist for detailsregress csat i.region


Interactions

Exercise 2: Regression, Categorical Predictors, &Interactions

Open the datafile, states.dta.1 Add on to the regression equation that you created in exercise 1 by

generating an interaction term and testing the interaction.2 Try adding a categorical variable to your regression (remember, it will

need to be dummy coded). You could use region or high25, orgenerate a new categorical variable from one of the continuousvariables in the dataset.


Exporting and saving results

Topic

1 Introduction



4 Interactions


6 Wrap-up



Saving and exporting regression tables

Usually when we’re running regression, we’ll be testing multiplemodels at a timeCan be difficult to compare resultsStata offers several user-friendly options for storing and viewingregression output from multiple modelsFirst, download the necessary packages:

* install outreg2 packagefindit outreg2



Saving and replaying

You can store regression model results in Stata

// fit two regression models and store the resultsregress csat expense income percent highestimates store Model1regress csat expense income percent high i.regionestimates store Model2




Stored models can be recalled

// Display Model1estimates replay Model1




Stored models can be compared

// Compare Model1 and Model2 coefficientsestimates table Model1 Model2



Exporting into Excel

Avoid human error when transferring coefficients into tablesExcel can be used to format publication-ready tables

outreg2 [Model1 Model2] using csatprediction.xls, replace


Wrap-up

Topic

1 Introduction



4 Interactions


6 Wrap-up


Wrap-up

Help Us Make This Workshop Better

Please take a moment to fill out a very short feedback formThese workshops exist for you–tell us what you need!ttp://tinyurl.com/StataRegressionFeedback


Wrap-up

Additional resources

training and consultingIQSS workshops:http://projects.iq.harvard.edu/rtc/filter_by/workshopsIQSS statistical consulting: http://rtc.iq.harvard.edu

Stata resourcesUCLA website: http://www.ats.ucla.edu/stat/Stata/Great for self-studyLinks to resources

Stata website: http://www.stata.com/help.cgi?contentsEmail list: http://www.stata.com/statalist/


http://projects.iq.harvard.edu/rtc/filter_by/workshops

http://rtc.iq.harvard.edu

http://www.ats.ucla.edu/stat/Stata/

http://www.stata.com/help.cgi?contents

http://www.stata.com/statalist/

stata statistics

Technology

introductionlaunch stata

todays stata script

scoresunivariate regression

zip unzip statastats

states data setuse states

data change directorycd

csat scores

expensesum csat expense