Quantitative Data Analysis Using STATA

Topics Covered in This Training

• Data Management in STATA: Importing data, cleaning data, creating new variables, and labeling variables.
• Descriptive Statistics:
  • Frequency tables
  • Summarizing data using mean, median, and standard deviation
  • Histograms, bar charts, and pie charts
• Hypothesis Testing:
  • Test of means (t-test)
  • Proportion tests
• Linear Regression Analysis:
  • Simple linear regression model
  • Multiple regression analysis
  • Interaction terms and their interpretation

Introduction

STATA is a statistical software designed for:

• Data manipulation or manage data,
• Graphics or data visualization, and
• Conduct data analysis.

STATA has different windows including:

• Command Window: Where commands are entered
• Results Window: Displays the output of commands
• Review Window: Shows a history of commands
• Variable Window: Lists the dataset variables
• Properties Window: Shows the properties of variables

Flavors of STATA

STATA 17 comes in different flavors:

• Basic edition (Stata/BE),
• Special edition (Stata/SE), and
• Multiprocessor (Stata/MP).

To explore the features of different STATA flavors, for instance, type the command:

help limits

This command shows the capacities of Stata/BE, Stata/MP, and Stata/SE.

Basic Commands

1. help

The help command provides access to STATA's built-in help documentation. It's useful when you need to learn more about a specific command or feature.

help summarize

This will display help for the summarize command.

2. search

The search command looks for specific information within the STATA help files and other resources. It's broader than help.

search regression

STATA will return results related to regression commands and topics.

3. pwd

The pwd command prints the current working directory. It's useful to know where STATA is currently looking for files.

pwd

This will output something like C:/Users/YourName/Documents/STATA/.

4. cd

Use cd to change the current working directory in STATA. This is important when you want STATA to look in a different folder for data files.

cd “C:\Users\user1\Desktop\STATA Training” 

5. import

STATA can import data from different file formats using the import command. For instance, to import an Excel file:

import excel "hh_91.xls", sheet("Sheet1") firstrow

This imports data from "Sheet1" of the Excel file and treats the first row as variable names.

If the data type is of csv format, the data can be imported using the import delimited command:

import delimited "hh_91.csv"

If the data type is, however, a STATA data type (.dta), the use command should be used to open the files.

use "hh_91.dta", clear

To view the summary of the data, the describe command is one option that can be utilized in STATA. The command produces a summary of the dataset in memory or of the data stored in a Stata-format dataset.

describe

6. edit

The edit command opens the Data Editor in edit mode, allowing you to modify data manually.

edit

This opens an interactive window to edit your dataset.

7. browse

The browse command opens the Data Editor in read-only mode. You can view, but not change the data.

browse

8. merge

To combine datasets based on one or more key variables, use merge. You can specify whether the merge is 1:1 or 1:many.

merge 1:1 id using "dataset2.dta"

This merges two datasets based on the variable id.

9. append

Use the append command to add observations from another dataset to the current one. It's like stacking datasets.

append using "dataset2.dta"

This adds the rows of dataset2 to the current dataset.

10. generate

The generate command creates new variables. For instance, to create a new variable for age in months:

generate age_months = agehead * 12

11. egen

The egen command is more advanced than generate and allows for group-level calculations or other complex operations.

egen mean_fsize = mean(famsize), by(villid)

This generates the mean family size for each village.

12. rename

To change the name of a variable, use rename.

rename weight samplewt

This renames the variable weight to samplewt.

13. drop

Removes variables or observations from the dataset.

drop if agehead > 50

This drops all observations where age is greater than 50.

14. keep

Keeps only the specified variables or observations.

keep if famsize <=15

This keeps only observations where income is greater than 1000.

15. clear

Clears the current dataset from memory. Useful when you want to load new data.

clear

16. encode

Converts string variables into numeric categorical variables.

encode gender, generate(gender_num)

This converts the string variable gender into a numeric variable gender_num.

17. decode

Converts numeric categorical variables back to strings.

decode gender_num, generate(gender_str)

This converts the numeric variable gender_num into a string variable gender_str.

18. replace

Replaces the values of existing variables.

replace income = 0 if income < 0

This replaces negative income values with 0.

19. label var

Attaches a label (up to 80 characters) to a variable.

label var sexhead "Sex of the household head"

This replaces the variable description.

20. label define

Creates a value label, which is a set of individual numeric values and their corresponding labels. For instance, for sexhead:

label define sex 1 "Male" 2 "Female"

This creates the value label named sex.

21. label val

Attaches a value label toa variable of your interest. For instance, for sexhead:

label val sexhead sex 

This attaches the value label named sex to the variable sexhead.

Descriptive Statistics

For categorical variables, tab1, tab2, and tabulate commands can be used to produce one way and two frequency tables.

tab1 produces a one-way tabulation for each variable specified after the command.

tab2 produces all possible two-way tabulations of the variables specified after the command.

Example: The following lines of codes produce different results:

tab1 sexhead vaccess 

tab2 sexhead vaccess 

tab2 sexhead vaccess, col 

tab2 sexhead vaccess, row 

Summarizing Numeric Variables

summarize agehead educhead 

Charting Numeric Variables

histogram agehead, frequency title("Distribution of Age") xlabel(15(5)75) ylabel(, grid) xtitle("Age (years)") ytitle("Frequency")

Box plots are useful in summarizing a numeric variable.

graph box age, title("Boxplot of Age") ytitle("Age")

If you want to produce a horizontal boxplot, you need to use the following command

graph hbox age, title("Boxplot of Age") xtitle("Age")

Density Plots

Density plots can also be generated for numeric variables, offering a smooth, continuous curve that eliminates the jagged appearance seen in the histogram above.

 kdensity exptot, title("Distribution of hh expenditures")

Density plots are helpful in making comparisons of distributions across populations or categories. The following graph displays a comparison of the total household expenditure across the sex of the household head

 twoway (kdensity exptot if sexhead==0) ///
(kdensity exptot if sexhead==1), ///
title("Distribution of hh expenditures" ///
 "by sex of the household head") ///
ytitle("Density") xtitle("HH expenditure") legend(label(1 "Female") label(2 "Male"))

Another useful way of comparison for distribution across groups is the use of boxplot.

graph hbox exptot, over(sexhead) title("Total Expenditure by Household Head's Sex") ylabel(, angle(0)) ytitle("Total Expenditure")

Charting Categorical Variables

graph bar (count), over(sex) title("Distribution of Sex of the Household Head") ytitle("Number") blabel(bar, position(outside) size(small))

Altenatively, a percentage distribution can be plotted on the bar chart instead of counts.

graph bar (percent), over(sex) title("Distribution of Sex of the Household Head") ytitle("Percent") blabel(bar, position(outside) size(small) format(%4.1f))

NOTE: After producing the chart, the chart needs to be saved as either as STATA graph file type or a picture. Shown below is the command you can use to save it as PNG file.

graph export "gbar.png", as(png) name("Graph")

A pie chart can also be produced for categorical variables. For instance, for sex of the households head:

graph pie, over(sex) title("Distribution of Sex") plabel(_all percent) legend(on)

Exporting Descriptive Results to MS Word

In STATA 17, table is a flexible command for creating tables of many types - tabulations, tables of summary statistics, tables of regression results, and more. The following lines of codes produce the mean and standard deviations of variables - age, level of education, family size, and land size of the household. The result is then exported to a word file named "MyResult.docx".

putdocx clear
putdocx begin 
table (var)()(), statistic(mean agehead educhead famsize hhland) ///
statistic(sd agehead educhead famsize hhland) nformat(%9.1f) 
collect style putdocx, title("Table 1: Descriptive Summary Measures")
putdocx collect 
putdocx save "MyResult", replace

It is also possible to export multiple summary tables to MS word.

putdocx clear
putdocx begin 
table (var)()(), statistic(mean agehead educhead famsize hhland) ///
statistic(sd agehead educhead famsize hhland) nformat(%9.1f) 
collect style putdocx, title("Table 1: Descriptive Summary Measures")
putdocx collect 

table (var)()(), statistic(fvfrequency sexhead vaccess)  ///
statistic(fvpercent sexhead vaccess) nformat(%9.1f)
collect recode result fvfrequency = Number fvpercent = Percent
collect layout (var) (result[Number Percent])
collect style putdocx, title("Table 2: Percent Distributions of  ... ")
putdocx collect 

putdocx save "MyResult", replace

Inferential Statistics - Estimation

In statistics, inference can be made in two ways. One way is estimation, point or interval estimation. Another way of making inference is by testing a hypothesis about a parameter.

Point and interval estimation for the mean are produced using the command ci means. For instance, for variables family size and household landholding size, the average estimated values - point and 95% confidence interval are estimated as follows:

ci mean famsize hhland

Similarly, you can produce the point and interval estimates for proportion of binary variables using the command ci proportion. Accordingly, the point and interval estimates for proportion of accessibility of village by road all year is produced using the following command.

ci proportion vaccess

The Wilson interval provides more accurate estimates, especially for proportions that are near 0 or 1. It accounts for the uncertainty in the estimate more effectively than the traditional normal approximation interval.

ci proportion vaccess

Inferential Statistics - Hypothesis Testing

Hypothesis testing helps in making informed decisions based on data. Researchers can determine whether to accept or reject a hypothesis about a population parameter. Hypothesis testing allows researchers to evaluate the strength of evidence against a null hypothesis. This is crucial in scientific research where conclusions must be supported by data.

For instance, we may test the hypthesis that the average age of household heads is 35.

ttest agehead==35

The ttest can also be used to test that a variable has the same mean within the two groups defined by groupvar. Similarly, assuming unpaired data, ttest can also be used to test that two variables have the same mean. For comparison of average annual expenditure for male and female headed households, we may specify the command as:

ttest exptot, by(sexhead)

To conduct a test of proportion in STATA, we use the prtest command. The command can be used for both one sample and two sample situations. For one sample prooprtion test, for instance - to test if proportion of households with access to road all year is 75%, we specify the command as:

prtest vaccess==0.75

For a two sample situation, comparing proportion of two groups - testing if an equal proportion of male and female headed households have access to road all year, specify the command as:

prtest vaccess, by(sexhead)

Correlation and Regression

Correlation and regression are both statistical techniques used to analyze relationships between variables, but they serve different purposes and provide different insights. Correlation measures the strength and direction of a linear relationship between two variables. Regression, on the other hand, models the relationship between a dependent variable and one or more independent variables. It aims to predict the dependent variable based on the independent variables.

For correlation analysis, either correlate OR pwcorr commands are used in STATA. For instance, the command below produces the correlation matrix for the specified variables.

correlate exptot agehead educhead famsize hhland hhasset

The pearson correlation coefficient assumes normality of variables. Normality tests such as swilk and sfrancia can be used to check normality of variables. In case the variables violate this assumption, an alternative coefficient should be produced such as Spearman's rank correlation coefficient or Kendal's tau.

spearman exptot agehead educhead famsize hhland hhasset

Simple and Multiple Linear Regression

Simple Linear Regression is ideal for understanding the relationship between two variables, while Multiple Linear Regression is used to explore relationships involving several predictors. Simple linear regression analyzes the relationship between two variables by fitting a linear equation to the observed data.

Multiple linear regression extends simple linear regression by analyzing the relationship between one dependent variable and two or more independent variables.

To fit a simple linear regression model for total expenditure with household asset as a predictor:

regress exptot hhasset

Multiple linear regression model can be fit as follows for the total annual household expenditure:

reg exptot i.sexhead agehead educhead famsize hhland hhasset i.vaccess