Getting Started in Frequencies, Crosstab,

Factor and Regression Analysis

(ver. 2.0 beta, draft)

Oscar Torres-Reyna
Data Consultant
[email protected]

http://dss.princeton.edu/training/
 Case study: intro
Search here in the
home page for this
dataset

Metadata

Codebook in two
formats

NOTE: When data is

not available in
Stata, you can
Datasets, two download the SPSS
formats: ACII and portable (*.por),
SPSS open it using SPSS
(available at the
Marginals DSS lab) and saving
it as Stata.
 Case study: frequencies
Distribution of electoral preferences and gender. According to the codebook
‘q5’ has the electoral question and ‘qa’ gender.
. tab q5 /*No weights*/

Q5. If the Presidential election were

held today and the candidates were
Barack Freq. Percent Cum.

Barack Obama and Joe Biden, the Democra 481 45.68 45.68
John McCain and Sarah Palin, the Republ 464 44.06 89.74 No weights
(VOL) Other/Neither 21 1.99 91.74
(VOL) Undecided/Don't know/no answer 87 8.26 100.00

Total 1,053 100.00

.
. tab q5 [aweight=weight] /*With weights*/

Q5. If the Presidential election were

held today and the candidates were
Barack Freq. Percent Cum.

Barack Obama and Joe Biden, the Democra 504.337749 47.90 47.90
John McCain and Sarah Palin, the Republ
(VOL) Other/Neither
449.487545
20.5570831
42.69
1.95
90.58
92.53
Using weights
(VOL) Undecided/Don't know/no answer 78.61762284 7.47 100.00

Total 1,053 100.00

.
. tab qa /*No weights*/

A. Gender
(DO NOT
ASK) Freq. Percent Cum.

Male
Female
493
560
46.82
53.18
46.82
100.00
No weights
Total 1,053 100.00
NOTE: At this point, it is strongly
recommended to open a log to keep a
.
. tab qa [aweight=weight] /*With weights*/
record of your work and to extract output,
type:
A. Gender
(DO NOT
ASK) Freq. Percent Cum. log using mywork.log
Male
Female
500.388396
552.611604
47.52
52.48
47.52
100.00
Using weights You could also open a do-file by typing
Total 1,053 100.00 doedit and copy your commands there.
 Case study: Electoral preferences by gender

. tab q5 qa [aw=weight], col row /*Electoral preferences by gender*/

Key

frequency
row percentage
column percentage

Q5. If the
Presidential election
were held today and A. Gender (DO NOT
the candidates were ASK)
Barack Male Female Total

Barack Obama and Joe 209.42078 294.91697 504.33775

41.52 58.48 100.00
41.85 53.37 47.90

John McCain and Sarah 252.9313 196.55625 449.487545

56.27 43.73 100.00
50.55 35.57 42.69

(VOL) Other/Neither 10.055739 10.5013441 20.557083

48.92 51.08 100.00
2.01 1.90 1.95

(VOL) Undecided/Don't 27.980574 50.637048 78.617623

35.59 64.41 100.00
5.59 9.16 7.47

Total 500.3884 552.6116 1,053

47.52 52.48 100.00
100.00 100.00 100.00
 Case study: Electoral preferences by age

. tab q5 f1 [aw=weight], col row /*Electoral preferences by age*/

Key

frequency
row percentage
column percentage

Q5. If the
Presidential election
were held today and
the candidates were F1. What is your age? F1. What is your age?
Barack 18-24 25-29 30-34 35-39 40-44 45-54 55-64 65 or old (VOL) No Total

Barack Obama and Joe 29.355119 26.435913 40.727272 46.595118 38.610873 129.51971 86.169373 102.39238 4.5319886 504.33775
5.82 5.24 8.08 9.24 7.66 25.68 17.09 20.30 0.90 100.00
77.57 50.53 40.92 51.51 37.59 51.68 50.32 43.21 40.00 47.90

John McCain and Sarah 6.2229886 22.18839 54.883049 36.825588 51.046351 99.992283 69.037199 104.76215 4.5295414 449.487545
1.38 4.94 12.21 8.19 11.36 22.25 15.36 23.31 1.01 100.00
16.44 42.42 55.14 40.71 49.69 39.90 40.31 44.21 39.98 42.69

(VOL) Other/Neither 0 0 2.1209543 2.4419715 4.51458561 3.1358789 2.7783459 5.56534701 0 20.557083

0.00 0.00 10.32 11.88 21.96 15.25 13.52 27.07 0.00 100.00
0.00 0.00 2.13 2.70 4.39 1.25 1.62 2.35 0.00 1.95

(VOL) Undecided/Don't 2.2672181 3.6879373 1.809561 4.5920698 8.5570854 17.952531 13.264407 24.219596 2.2672179 78.617623
2.88 4.69 2.30 5.84 10.88 22.84 16.87 30.81 2.88 100.00
5.99 7.05 1.82 5.08 8.33 7.16 7.75 10.22 20.01 7.47

Total 37.845325 52.312241 99.540836 90.454747 102.7289 250.600407 171.24932 236.93948 11.328748 1,053
3.59 4.97 9.45 8.59 9.76 23.80 16.26 22.50 1.08 100.00
100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00
 Case study: Electoral preferences by educational attainment

. tab q5 f4 [aw=weight], col row /*Electoral preferences by education*/

Key

frequency
row percentage
column percentage

Q5. If the
Presidential election
were held today and
the candidates were F4. What is the highest grade of schooling that you've completed?
Barack 8th grade Some high High scho Some coll College g Postgradu (VOL) No Total

Barack Obama and Joe 2.2991619 3.883265 81.589679 113.53524 169.39657 130.23545 3.3983797 504.33775
0.46 0.77 16.18 22.51 33.59 25.82 0.67 100.00
41.27 33.65 44.32 45.42 44.89 59.52 60.03 47.90

John McCain and Sarah 3.2718681 6.1159475 76.7484051 116.69213 170.30303 74.093841 2.2623235 449.487545
0.73 1.36 17.07 25.96 37.89 16.48 0.50 100.00
58.73 53.00 41.69 46.68 45.13 33.86 39.97 42.69

(VOL) Other/Neither 0 0 3.7389017 3.382658 9.8911577 3.5443656 0 20.557083

0.00 0.00 18.19 16.45 48.12 17.24 0.00 100.00
0.00 0.00 2.03 1.35 2.62 1.62 0.00 1.95

(VOL) Undecided/Don't 0 1.5397725 22.004128 16.367784 27.7818421 10.924096 0 78.617623

0.00 1.96 27.99 20.82 35.34 13.90 0.00 100.00
0.00 13.34 11.95 6.55 7.36 4.99 0.00 7.47

Total 5.57103 11.538985 184.08111 249.97781 377.3726 218.79775 5.6607032 1,053

0.53 1.10 17.48 23.74 35.84 20.78 0.54 100.00
100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00
 Case study: Electoral preferences by income

. tab q5 f13 [aw=weight], col row /*Electoral preferences by income*/

Key

frequency
row percentage
column percentage

Q5. If the
Presidential election
were held today and
the candidates were F13. Finally, just for classification purposes, was your total family income bef
Barack Less than $20,000 t $35,000 t $50,000 t $75,000 t $100,000 or $150,0 (VOL) No Total

Barack Obama and Joe 37.525195 51.14097 72.715849 122.78749 59.632459 69.732723 51.1129092 39.690155 504.33775
7.44 10.14 14.42 24.35 11.82 13.83 10.13 7.87 100.00
60.42 49.40 48.59 57.51 39.05 46.16 44.46 37.63 47.90

John McCain and Sarah 18.630762 39.764056 64.4115908 69.827216 86.023642 68.843117 54.640308 47.346852 449.487545
4.14 8.85 14.33 15.53 19.14 15.32 12.16 10.53 100.00
30.00 38.41 43.04 32.71 56.34 45.57 47.53 44.88 42.69

(VOL) Other/Neither 1.5060026 .88321203 3.2060684 2.5018142 2.1243815 3.0806277 2.200355 5.0546217 20.557083
7.33 4.30 15.60 12.17 10.33 14.99 10.70 24.59 100.00
2.42 0.85 2.14 1.17 1.39 2.04 1.91 4.79 1.95

(VOL) Undecided/Don't 4.4480018 11.739914 9.3136182 18.37691 4.9181423 9.409895 7.01703324 13.3941079 78.617623
5.66 14.93 11.85 23.38 6.26 11.97 8.93 17.04 100.00
7.16 11.34 6.22 8.61 3.22 6.23 6.10 12.70 7.47

Total 62.109961 103.52815 149.64713 213.49343 152.69863 151.06636 114.97061 105.48574 1,053
5.90 9.83 14.21 20.27 14.50 14.35 10.92 10.02 100.00
100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00
 Case study: Electoral preferences by employment status

. tab q5 f8 [aw=weight], col row /*Electoral preferences by employment status*/

Key

frequency
row percentage
column percentage

Q5. If the
Presidential election
were held today and
the candidates were f8
Barack Employed Employed Laid off Retired Student Homemaker Something (VOL) No Total

Barack Obama and Joe 263.30095 36.9693237 17.692466 125.50328 15.486465 16.644394 24.6988275 4.0420475 504.33775
52.21 7.33 3.51 24.88 3.07 3.30 4.90 0.80 100.00
47.30 52.01 67.23 46.14 82.02 27.25 60.77 64.12 47.90

John McCain and Sarah 252.31686 25.723928 6.1500438 112.5963 1.1268505 37.187532 12.123702 2.2623235 449.487545
56.13 5.72 1.37 25.05 0.25 8.27 2.70 0.50 100.00
45.33 36.19 23.37 41.39 5.97 60.89 29.83 35.88 42.69

(VOL) Other/Neither 11.498793 1.6530186 0 6.1126834 0 0 1.2925883 0 20.557083

55.94 8.04 0.00 29.74 0.00 0.00 6.29 0.00 100.00
2.07 2.33 0.00 2.25 0.00 0.00 3.18 0.00 1.95

(VOL) Undecided/Don't 29.558151 6.7386098 2.4747578 27.814172 2.2672181 7.2399743 2.52474 0 78.617623
37.60 8.57 3.15 35.38 2.88 9.21 3.21 0.00 100.00
5.31 9.48 9.40 10.22 12.01 11.85 6.21 0.00 7.47

Total 556.67476 71.08488 26.3172676 272.02643 18.8805338 61.0719 40.639858 6.304371 1,053
52.87 6.75 2.50 25.83 1.79 5.80 3.86 0.60 100.00
100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00
 Case study: Testing for associations (preparing the data)
Before running any test we need to prepare the data by setting to missing any non-valid response (like
“don’t know/no answer/not sure”) unless is relevant to the question. It is important to ‘clean’ the variables
for the tests to be as accurate as possible. For demographics we will remove non-response items. Here are
a series of commands per variable (columns) to prepare some variables for you to run on your own.

Description Age Education Income Employment Gender

creating a new
gen age=f1 gen educ=f4 gen income=f13 gen employ=f8 gen gender=qa
variable

exploring the new

tab age tab educ tab income tab employ tab gender
variable

checking for labels

labelbook f1 labelbook f4 labelbook f13 labelbook f8 labelbook qa
from original variable

assigning labels to label value

label value age f1 label value educ f4 label value income f13 label value employ f8
new variable gender qa

exploring the new

tab age tab educ tab income tab employ tab gender
variable

setting no response to replace age=. if replace income=. if

replace educ=. if educ==8 replace employ=. if employ==8
missing age>8 income==8

label variable age label variable educ "Educational label variable income label variable employ
adding variable labels
"Age" attainment" "Family income" "Employment status"

exploring the new

tab age tab educ tab income tab employ
variable
 Case study: Testing for associations (preparing the data –cont.)
Here is an easy way to do it by using the command clonevar in Stata.

Description Age Education Income Employment Gender

creating a new clonevar clonevar clonevar clonevar

clonevar employ=f8
variable age=f1 educ=f4 income=f13 gender=qa

exploring the new

tab age tab educ tab income tab employ tab gender
variable

setting no
replace age=. replace educ=. replace income=. if replace employ=. if
response to
if age>8 if educ==8 income==8 employ==8
missing

exploring the new

tab age tab educ tab income tab employ
variable
 Case study: testing for associations

To find whether there is some association between demographics and electoral

preferences we can use chi-square but first we need to ‘clean’ the electoral variable (q5).
Lets create a new variable ‘elec’ from ‘q5’. We will use recode for this, type:

Original variable Value 2=2 with label in quotes New variable, name in parenthesis

Value 1=1 with label in quotes Values 3, 4 & 8 = 3 with label in quotes

. tab elec Labels are saved as ‘elec’

RECODE of q5 (Q5. If
the Presidential
election were held
today and the
candidate Freq. Percent Cum.

Obama/Biden 481 45.68 45.68

Here is the new variable
McCain/Palin 464 44.06 89.74
Undecided/DK/NA/Other 108 10.26 100.00

Total 1,053 100.00

We use the ‘nofreq’ option after comma since we

are not interested on the crosstabulations but rather
on the tests. We can see that gender, education,
income and employment status are somehow
associated with electoral preferences. Age does not
seem to have any association.
 Case study: descriptive statistics

When you have continuous data you need to use descriptive statistics. To start exploring
this option you can use the summarize command which provides first look at the data
(number of observations, mean, standard deviation, minimum and maximum values).
Lets take a look at the battery of questions in q8.

The questions ask for answers

between 0 and 100. The
maximum value 999 represents
“Not answer/Not sure”
response. The mean and
standard deviation factor in the
999 therefore biasing the mean
and sd. so we need to set 999
to missing so the values go
from 0 to 100.

Here 999 is set to missing and

we have correct statistics (see
the slides on ‘preparing the
data’ to do this). For
presentation purposes we
won’t use weights here.
 Case study: descriptive statistics

To get more than the mean and sd you can use tabstat which offers several
options (type help tabstat for more details). Notice we use weights here.
In these series of questions ‘0’ means ‘unfavorable’ and ‘100’ favorable.
. tabstat x8a x8b x8c x8d x8e x8f x8g x8h x8i x8j, s(mean median sd var count range min max)

stats x8a x8b x8c x8d x8e x8f x8g x8h x8i x8j

mean 55.83044 54.43365 55.87257 49.39961 40.33595 56.01527 55.93845 34.61905 53.96314 60.41824
p50 60 55 60 50 50 60 55 30 55 65
sd 35.31804 31.28831 31.18157 35.33493 24.2347 26.50595 22.22173 31.2718 23.95454 22.56533
variance 1247.364 978.9581 972.2905 1248.557 587.3207 702.5655 493.8053 977.9253 573.82 509.194
N 1038 1040 1028 1036 1018 982 991 1050 1031 1009
range 100 100 100 100 100 100 100 100 100 100
min 0 0 0 0 0 0 0 0 0 0
max 100 100 100 100 100 100 100 100 100 100

Here is a description of each variable

. describe x8*

storage display value

variable name type format label variable label

x8a float %9.0g Obama

x8b float %9.0g McCain
x8c float %9.0g Biden
x8d float %9.0g Palin
x8e float %9.0g Congress
x8f float %9.0g Congressman
x8g float %9.0g Supreme court
x8h float %9.0g Pres. Bush
x8i float %9.0g State gov
x8j float %9.0g Local gov
 Case study: descriptive statistics
Lets explore a combination of commands to get more info out of your data. We will check
out the battery of questions in q25

The questions ask for answers

between 0 and 10 (see the
codebook) . The maximum
value 99 (below) represents
“Not answer/Not sure”
response.

The mean and standard

deviation factor in the 99
therefore biasing the mean and
sd. so we need to set 99 to
missing so the values go from
0 to 10 (see the slides on
‘preparing the data’ to do this).
 Case study: descriptive statistics
Here some descriptive statistics for q25 where a value of ‘0’ or ’1’ represents ‘strongly
oppose’ and value of ‘9’ or ’10’ represents ‘strongly favor’.

Here we use the combination tab/sum to explore a

response to a third variable (usually continuous) in a
crosstabulation. We are looking at the mean value of x25c
(‘govt should provide health care’) by electoral preference
and gender. For example, male Obama supporters tend to
support government providing health care who can’t afford
it (mean of 8.19). On the contrary, those who are male and
prefer McCain tend to disagree (with a mean score of 2.76)
 Case study: dummies
The quickest way to generate dummy variables is by using a combination of tab/gen
command. Here is an example
 Case study: factor analysis
Factor analysis is a data reduction technique. Question 8 has a battery of questions
evaluating favorability levels for different candidates/politicians

Principal-components factoring
Variables
Since the sum of eigenvalues
= total number of variables.
. factor x8a x8b x8c x8d x8e x8f x8g x8h x8i x8j, pcf
Proportion indicate the relative
Total variance accounted by (obs=897) weight of each factor in the
each factor. The sum of all total variance. For example,
Factor analysis/correlation Number of obs = 897 4.109/10=0.4109. The first
eigenvalues = total number of Method: principal-component factors Retained factors = 2
variables. Rotation: (unrotated) Number of params = 19 factor explains 41% of the total
When negative, the sum of variance
eigenvalues = total number of Factor Eigenvalue Difference Proportion Cumulative
factors (variables) with positive
Factor1 4.10910 1.89154 0.4109 0.4109
eigenvalues. Factor2 2.21756 1.35886 0.2218 0.6327
Kaiser criterion suggests to Factor3 0.85870 0.12199 0.0859 0.7185
Factor4 0.73671 0.15331 0.0737 0.7922
retain those factors with Factor5 0.58340 0.05168 0.0583 0.8505 Cumulative shows the amount
eigenvalues equal or higher Factor6 0.53172 0.13910 0.0532 0.9037
of variance explained by n+(n-
than 1. Factor7 0.39262 0.11864 0.0393 0.9430
Factor8 0.27398 0.11808 0.0274 0.9704 1) factors. For example, factor
Factor9 0.15591 0.01559 0.0156 0.9860 1 and factor 2 account for 63%
Factor10 0.14031 . 0.0140 1.0000
of the total variance.
LR test: independent vs. saturated: chi2( 45) = 4884.51 Prob>chi2 = 0.0000

Factor loadings (pattern matrix) and unique variances

Difference between one
eigenvalue and the next.
Variable Factor1 Factor2 Uniqueness

x8a -0.9046 0.1045 0.1709

x8b 0.8586 0.2150 0.2165
x8c -0.8531 0.1799 0.2399 Uniqueness is the variance that is ‘unique’
x8d 0.9180 0.1434 0.1367
x8e -0.4759 0.5533 0.4674 to the variable and not shared with other
x8f -0.1691 0.6717 0.5202 variables. It is equal to 1 – communality
x8g 0.2197 0.5555 0.6432
x8h 0.8225 0.2936 0.2373
(variance that is shared with other
x8i -0.0373 0.7252 0.4728 variables). For example, 64% of the
x8j -0.0425 0.6554 0.5686 variance in ‘x8g’ is not share with other
variables in the overall factor model. On the
contrary ‘x8a’ has low variance not
Factor loadings are the weights and correlations between each variable and the factor. The accounted by other variables (17%). Notice
higher the load the more relevant in defining the factor’s conceptual meaning. A negative that the greater ‘uniqueness’ the lower the
value indicates an inverse impact on the factor. Here, two factors are retained because relevance of the variable in the factor model.
both have eigenvalues over 1. It seems that ‘x8b’, ‘x8d’ and ‘x8h’ define factor1, and ‘x8f’,
and ‘x8i’ define factor2.
 Case study: factor analysis

Factor analysis is a data reduction technique. Question 8 has a battery of questions evaluating
favorability levels for different candidates/politicians

By default the rotation is varimax which . rotate

produces orthogonal factors. This means
that factors are not correlated to each other. Factor analysis/correlation Number of obs = 897
Method: principal-component factors Retained factors = 2
This setting is recommended when you Rotation: orthogonal varimax (Kaiser off) Number of params = 19
want to identify variables to create indexes
or new variables without inter-correlated
components Factor Variance Difference Proportion Cumulative

Factor1 4.08288 1.83911 0.4083 0.4083

Same description as in the previous slide Factor2 2.24377 . 0.2244 0.6327
with new composition between the two LR test: independent vs. saturated: chi2( 45) = 4884.51 Prob>chi2 = 0.0000
factors. Still both factors explain 63% of the
total variance observed. Rotated factor loadings (pattern matrix) and unique variances

The pattern matrix here offers a clearer Variable Factor1 Factor2 Uniqueness
picture of the relevance of each variable in x8a -0.8860 0.2103 0.1709
the factor. x8b 0.8780 0.1124 0.2165
x8c -0.8260 0.2790 0.2399
x8d 0.9285 0.0343 0.1367
x8e -0.4075 0.6055 0.4674
x8f -0.0888 0.6869 0.5202
x8g 0.2836 0.5257 0.6432
x8h 0.8513 0.1947 0.2373
x8i 0.0483 0.7245 0.4728
x8j 0.0350 0.6559 0.5686

Factor rotation matrix

Factor1 Factor2
This is a correlation matrix between factor1
and factor2. Factor1 0.9930 -0.1177
Factor2 0.1177 0.9930

NOTE: If you want the factors to be correlated (oblique rotation) you need to use the option promax after rotate:
rotate, promax
Type help rotate for details.
 Case study: factor analysis, step 3 (predict)
To create the new variables, after factor, rotate you type predict.

predict x8f1 x8f2 /*Or whatever name you prefer to identify the factors*/

. predict x8f1 x8f2

(regression scoring assumed)

Scoring coefficients (method = regression; based on varimax rotated factors)

Variable Factor1 Factor2

x8a -0.21306 0.07271

x8b 0.21892 0.07169
x8c -0.19662 0.10498
x8d 0.22947 0.03792
x8e -0.08565 0.26140
x8f -0.00521 0.30564
x8g 0.08259 0.24245
x8h 0.21436 0.10790
x8i 0.02947 0.32580
x8j 0.02453 0.29473

These are the regression coefficients used to estimate

the individual scores (per case/row)

We reduced all eight variables to two: x8f1 and x8f2. There is another way to use these results. We could
create indexes out of each cluster of variables. For example, ‘x8b’, ‘x8d’ and ‘x8h’ define the first factor.
You could aggregate these to create a new variable to measure ‘Republican favorability’. The second
factor is defined by ‘x8e’, ‘x8f’, x8i’ and ‘x8j’ related to ‘government institutions’. Since all variables are in
the same valence (go from 0 to 100), we can create the two new variables as

gen repubfav = (x8b + x8d + x8h)/3

gen govinst = (x8e + x8f + x8i + x8j)/4
 Case study: regression
We use the command regress to run a regression

regress x8a gender age educ income x25*, robust

. regress x8a gender age educ income x25*, robust

Linear regression Number of obs = 857

F( 14, 842) = 138.68
Prob > F = 0.0000
R-squared = 0.6114
Root MSE = 22.13

Robust
x8a Coef. Std. Err. t P>|t| [95% Conf. Interval]

gender 1.085681 1.524235 0.71 0.476 -1.906064 4.077427

age -.0954027 .4441548 -0.21 0.830 -.9671832 .7763779
educ 1.570695 .8151773 1.93 0.054 -.0293229 3.170713
income -.2996345 .4764621 -0.63 0.530 -1.234827 .6355583
x25a 1.101605 .2762611 3.99 0.000 .5593636 1.643846
x25b .6041541 .2659564 2.27 0.023 .0821388 1.126169
x25c 2.749842 .3712377 7.41 0.000 2.021182 3.478502
x25d -.1274084 .3054922 -0.42 0.677 -.7270241 .4722072
x25e -.2741189 .2758408 -0.99 0.321 -.8155351 .2672973
x25f -.9597492 .3174276 -3.02 0.003 -1.582792 -.3367069
x25g 1.201146 .3624039 3.31 0.001 .4898251 1.912467
x25h -2.622509 .3181912 -8.24 0.000 -3.24705 -1.997968
x25i -.6518584 .3177172 -2.05 0.041 -1.275469 -.0282476
x25j .699863 .3073602 2.28 0.023 .0965809 1.303145
x25f1 (dropped)
x25f2 (dropped)
x25f3 (dropped)
_cons 39.59818 7.345718 5.39 0.000 25.18011 54.01625
 Case study: regression
Here is another example

regress x8b gender age educ income x25*, robust

. regress x8b gender age educ income x25*, robust

Linear regression Number of obs = 857

F( 14, 842) = 70.66
Prob > F = 0.0000
R-squared = 0.4955
Root MSE = 22.135

Robust
x8b Coef. Std. Err. t P>|t| [95% Conf. Interval]

gender 2.568956 1.529457 1.68 0.093 -.4330398 5.570951

age -.3590177 .4220541 -0.85 0.395 -1.187419 .469384
educ 2.394501 .8376223 2.86 0.004 .7504277 4.038573
income .7567806 .5004008 1.51 0.131 -.2253989 1.73896
x25a -.4245393 .2513435 -1.69 0.092 -.9178727 .068794
x25b -.5100364 .2616189 -1.95 0.052 -1.023538 .0034653
x25c -1.546259 .3302899 -4.68 0.000 -2.194547 -.8979706
x25d -.0041063 .2839938 -0.01 0.988 -.5615252 .5533125
x25e -.5360159 .2764522 -1.94 0.053 -1.078632 .0066005
x25f 1.08052 .3298975 3.28 0.001 .4330022 1.728038
x25g -.2805339 .3361083 -0.83 0.404 -.9402424 .3791746
x25h 3.539997 .3070789 11.53 0.000 2.937267 4.142727
x25i .5077791 .3273211 1.55 0.121 -.134682 1.15024
x25j -.0397483 .2948785 -0.13 0.893 -.6185315 .5390349
x25f1 (dropped)
x25f2 (dropped)
x25f3 (dropped)
_cons 28.87047 7.224851 4.00 0.000 14.68964 43.0513
 Case study: regression (exporting results)
Use the outreg2 command to export the
output in a journal-paper like presentation. Run
outreg2 after each regression as follows

regress x8a gender age educ income x25*, robust

outreg2 using model, bdec(2) tdec(2) rdec(2) adec(2)

alpha(0.001, 0.01, 0.05) addstat(Adj. R-squared, e(r2_a))
word

Click here to see the document

regress x8b gender age educ income x25*, robust

outreg2 using model, bdec(2) tdec(2) rdec(2) adec(2)

alpha(0.001, 0.01, 0.05) addstat(Adj. R-squared, e(r2_a))
word append

Click here to see the document

 Case study: do-file (part1)
log using workshop.log /*Preparing income variable*/
gen income=f13
/*Distribution of electoral preferences (frequencies)*/ tab income
labelbook f13
tab q5 /*No weights*/ label value income f13
tab q5 [aweight=weight] /*With weights*/ tab income
replace income=. if income==8
tab qa /*No weights*/ label variable income "Family income"
tab qa [aweight=weight] /*With weights*/ tab income

/*Electoral preferences by some demographics*/ /*Preparing employment variable*/

gen employ=f8
tab q5 qa [aw=weight], col row /*Electoral preferences by gender*/ tab employ
tab q5 f1 [aw=weight], col row /*Electoral preferences by age*/ labelbook f8
tab q5 f4 [aw=weight], col row /*Electoral preferences by education*/ label value employ f8
tab q5 f13 [aw=weight], col row /*Electoral preferences by income*/ tab employ
tab q5 f8 [aw=weight], col row /*Electoral preferences by employment status*/ replace employ=. if employ==8
label variable employ "Employment status"
/*Preparing age variable*/ tab employ
gen age=f1
tab age /*Preparing gender variable*/
labelbook f1 gen gender=qa
label value age f1 tab gender
tab age labelbook qa
replace age=. if age>8 label value gender qa
label variable age "Age" tab gender
tab age
/*Recoding electoral question*/
/*Preparing education variable*/ recode q5 (1=1 "Obama/Biden") (2=2 "McCain/Palin") (3 4 8=3
"Undecided/DK/NA/Other"), gen(elec) label(elec)
gen educ=f4 tab q5
tab educ tab elec
labelbook f4
label value educ f4 /*Testing for associations*/
tab educ tab elec gender, nofreq chi2
replace educ=. if educ==8 tab elec age, nofreq chi2
label variable educ "Educational attainment" tab elec educ, nofreq chi2
tab educ tab elec income, nofreq chi2
tab elec employ, nofreq chi2
/*Factor, data preparation*/ Case study: do-file (part 2)
gen x8a = q8a
gen x8b = q8b /*Descriptive statistics*/
gen x8c = q8c
gen x8d = q8d tabstat q8a x8a q8b x8b, s(mean)
gen x8e = q8e
gen x8f = q8f tabstat x8a x8b x8c x8d x8e x8f x8g x8h x8i x8j, s(mean median sd var count
gen x8g = q8g range min max)
gen x8h = q8h
gen x8i = q8i describe x8*
gen x8j = q8j
/* One more factor example */
replace x8a = . if x8a>100
replace x8b = . if x8b>100 gen x25a = q25a
replace x8c = . if x8c>100 gen x25b = q25b
replace x8d = . if x8d>100 gen x25c = q25c
replace x8e = . if x8e>100 gen x25d = q25d
replace x8f = . if x8f>100 gen x25e = q25e
replace x8g = . if x8g>100 gen x25f = q25f
replace x8h = . if x8h>100 gen x25g = q25g
replace x8i = . if x8i>100 gen x25h = q25h
replace x8j = . if x8j>100 gen x25i = q25i
gen x25j = q25j
label variable x8a "Obama"
label variable x8b "McCain" replace x25a = . if x25a>10
label variable x8c "Biden" replace x25b = . if x25b>10
label variable x8d "Palin" replace x25c = . if x25c>10
label variable x8e "Congress" replace x25d = . if x25d>10
label variable x8f "Congressman" replace x25e = . if x25e>10
label variable x8g "Supreme court" replace x25f = . if x25f>10
label variable x8h "Pres. Bush" replace x25g = . if x25g>10
label variable x8i "State gov" replace x25h = . if x25h>10
label variable x8j "Local gov" replace x25i = . if x25i>10
replace x25j = . if x25j>10
/*Running factor analysis */

factor x8a x8b x8c x8d x8e x8f x8g x8h x8i x8j, pcf
rotate
predict x8f1 x8f2

gen repubfav = (x8b + x8d + x8h)/3

gen govinst = (x8e + x8f + x8i + x8j)/4
 Case study: do-file (part 3)
label variable x25a "A woman should be able to get an abortion if she wants one in the first three months of pregnancy, no matter what the reason"
label variable x25b "Gay couples should be allowed to marry, giving them full legal rights of married couples"
label variable x25c "The government should provide health care coverage to all citizens who can’t afford it, even if it means higher taxes"
label variable x25d "Government regulation of financial institutions should be greatly increased"
label variable x25e "The government should have let financial institutions that got into trouble over bad mortgage debt go out of business rather than
trying to rescue them"
label variable x25f "The government should allow offshore drilling for oil and gas in the waters off the U.S. coast "
label variable x25g "Congress should pass stricter laws to protect the environment and reduce global warming, even if the economic costs are high"
label variable x25h "Our troops should stay in Iraq without a timetable for withdrawal until the Iraqi government is stable"
label variable x25i "Government should cut taxes on businesses to help the economy"
label variable x25j "The government should help people who can’t afford their mortgage payments by suspending foreclosures until the economy has
improved"

factor x25a x25b x25c x25d x25e x25f x25g x25h x25i x25j, pcf
rotate
predict x25f1 x25f2 x25f3

/*Regression*/

regress x8a gender age educ income x25*, robust

regress x8b gender age educ income x25*, robust
 Exploring data:
annotated output
 Exploring data: frequencies (intro)
Frequency refers to the number of times a value is repeated. Frequencies are usually used
to analyze categorical data. The tables below are frequency tables. Values are in ascending
order. Use the command tab (type help tab for more details)

‘Freq.’ provides a raw count of each value. In this

case 10 students for each major.
‘Percent’ gives the relative frequency for each
value. For example, 33.33% of the students in this
group are econ majors.
‘Cum.’ is the cumulative frequency in ascending
order of the values. For example, 66.67% of the
students are econ or math majors.

‘Freq.’ Here 6 students read the newspaper 3 days

a week, 9 students read it 5 days a week.
‘Percent’. Those who read the newspaper 3 days a
week represent 20% of the sample, 30% of the
students in the sample read the newspaper 5 days
a week.
‘Cum.’ 66.67% of the students read the newspaper
3 to 5 days a week.
 Exploring data: crosstabs
Also known as contingency tables, crosstabs help you to analyze the relationship between two or more
variables (mostly categorical). Below is a crosstab between the variable ‘ecostatu’ and ‘gender’. We use
the command tab (with two variables to make the crosstab).
Options ‘col’, ‘row’ gives you the column The first value in a cell tells you the number
and row percentages. of observations for each xtab. In this case,
90 respondents are ‘male’ and said that the
var1 var2 economy is doing ‘very well’, 59 are
‘female’ and believe the economy is doing
‘very well’
The second value in a cell gives you row
percentages for the first variable in the
xtab. Out of those who think the economy
is doing ‘very well’, 60.40% are males and
39.60% are females.

The third value in a cell gives you column

percentages for the second variable in the
xtab. Among males, 14.33% think the
economy is doing ‘very well’ while 7.92% of
females have the same opinion.

You can use tab1 for multiple frequencies or tab2 to run

all possible crosstabs combinations. Type help tab for
further details.
 Exploring data: crosstabs (a closer look)
You can use crosstabs to compare responses among categories in relation to aggregate
responses. In the table below we compare male and female responses vs. the national
aggregate.
As a rule-of-thumb, a margin of error of ±4 percentage points can be
used to indicate a significant difference (some use ±3).
For example, rounding up the percentages, 11% (10.85) answer
‘very well’ at the national level. With the margin of error, this gives a
range roughly between 7% and 15%, anything beyond this range
could be considered significantly different (remember this is just an
approximation). It does not appear to be a significant bias between
males and females for this answer.
In the ‘fairly well’ category we have 49%, with range between 45%
and 53%. The response for males is 54% and for females 45%. We
could say here that males tend to be a bit more optimistic on the
economy and females tend to be a bit less optimistic.
If we aggregate responses, we could get a better picture. In the table
below 68% of males believe the economy is doing well (comparing to
60% at the national level, while 46% of females thing the economy is
bad (comparing to 39% aggregate). Males seem to be more
optimistic than females.

recode ecostatu (1 2 = 1 "Well") (3 4 = 2 "Bad") (5 6=3 "Not sure/ref"), gen(ecostatu1) label(eco)

 Exploring data: crosstabs (test for associations)
To see whether there is a relationship between two variables you can choose a number of
tests. Some apply to nominal variables some others to ordinal. I am running all of them here
for presentation purposes.

Likelihood-ratio χ2(chi-square) Goodman & Kruskal’s γ (gamma)

X2(chi-square) Cramer’s V Kendall’s τb (tau-b)

– For nominal data use chi2, lrchi2, V

– For ordinal data use gamma and taub
Fisher’s exact test
– Use exact instead of chi2 when
frequencies are less than 5 across the
table.

X2(chi-square) tests for relationships between variables. The null

hypothesis (Ho) is that there is no relationship. To reject this we
need a Pr < 0.05 (at 95% confidence). Here both chi2 are
significant. Therefore we conclude that there is some relationship
between perceptions of the economy and gender

Cramer’s V is a measure of association between two nominal

variables. It goes from 0 to 1 where 1 indicates strong association
(for rXc tables). In 2x2 tables, the range is -1 to 1. Here the V is
0.15, which shows a small association.
Gamma and taub are measures of association between two ordinal
variables (both have to be in the same direction, i.e. negative to
positive, low to high). Both go from -1 to 1. Negative shows inverse
relationship, closer to 1 a strong relationship. Gamma is
recommended when there are lots of ties in the data. Taub is
recommended for square tables.
Fisher’s exact test is used when there are very few cases in the
cells (usually less than 5). It tests the relationship between two
variables. The null is that variables are independent. Here we
reject the null and conclude that there is some kind of relationship
between variables
 Exploring data: descriptive statistics
For continuous data we use descriptive statistics. These statistics are a collection of
measurements of two things: location and variability. Location tells you the central value of
your variables (the mean is the most common measure of this) . Variability refers to the
spread of the data from the center value (i.e. variance, standard deviation). Statistics is
basically the study of what causes such variability. We use the command tabstat to get
these stats (the ‘s’ after the comma means ‘statistics’).

•The mean is the sum of the observations divided by the total number of observations.
•The median (p50 in the table above) is the number in the middle . To get the median you have to order
the data from lowest to highest. If the number of cases is odd the median is the single value, for an even
number of cases the median is the average of the two numbers in the middle.
•The standard deviation is the squared root of the variance. Indicates how close the data is to the mean.
Assuming a normal distribution, 68% of the values are within 1 sd from the mean, 95% within 2 sd and
99% within 3 sd
•The variance measures the dispersion of the data from the mean. It is the simple mean of the squared
distance from the mean.
•Count (N in the table) refers to the number of observations per variable.
•Range is a measure of dispersion. It is the difference between the largest and smallest value, max – min.
•Min is the lowest value in the variable.
•Max is the largest value in the variable.
 Exploring data: regression (what to look for)
Lets run the regression:
regress x8a gender age educ income, robust
This is the p-value of the model. It
Dependent Independent To control for indicates the reliability of X to
variable (Y) variables (X) heteroskedasticity predict Y. Usually we need a p-
value lower than 0.05 to show a
statistically significant relationship
between X and Y.

R-square shows the amount of

variance of Y explained by X. In
this case the model explains 5%
of the variance in x8a.

x8a = 25.88 + 9.5*gender – 0.8*age

+ 6.9*educ -2.3*income
These are two-tail p-values for each coefficient. It tests the
The t-values test the hypothesis that the coefficient is hypothesis that the coefficient is different from 0. To reject this, the p-
different from 0. To reject this, you need a t-value greater value has to be lower than 0.05 (you could choose also an alpha of
than 1.96 (95% confidence). You can get the t-values by 0.01). In this case, only “age” does not seem to be significant.
dividing the coefficient by its standard error. The t-values
also show the importance of a variable in the model. In
this case, educ is the most important.
 Exploring data: regression, publishing regression output (outreg2)
Once you define your final model, you can export your regression results using either your log file or the option outreg2. For the log
you just open it using any word processor and copy-and-paste the regression table into excel or word. The command outreg2 gives
you the type of presentation you see in scholar’s papers. Let’s say the final regression is
regress csat percent percent2 high
After running the regression type the following if you want to export the results to excel*
outreg2 using results, bdec(2) tdec(2) rdec(2) adec(2) alpha(0.001, 0.01, 0.05) addstat(Adj. R-
squared, e(r2_a)) excel
Or this if you want to export to word
outreg2 using results, bdec(2) tdec(2) rdec(2) adec(2) alpha(0.001, 0.01, 0.05) addstat(Adj. R-
squared, e(r2_a)) word
You will see this in Stata’s output window
For excel

Click here to see the output, a excel/word window will open

For word

Name of Set # of Set # of decimals Include some additional statistic, in this

the file for decimals for for added statistics case adj. R-sqr. You can select any
the output auxiliary (addstat option) statistics on the return lists (e-class, r-
statistics class or s-class). After running the
regression type ereturn list for a list
Click on seeout Set # of Set # of Levels of
of available statistics.
to browse the decimals decimals significance
results for for the R2
coefficients

Type help outreg2 for more details. If you do not see outreg2, you may have to install it by typing ssc install outreg2. If this does not work type
findit outreg2, select from the list and click “install”.
Note: If you get the following error message (when you use the option append or replace it means that you need to close the excel/word window.

*See the following document for some additional info/tips http://www.fiu.edu/~tardanic/brianne.pdf

 Exploring data: regression, publishing regression output (outreg2)
This is how the output would like (you will still need to do some additional editing):
In excel In word

You can add more models to compare. Lets say you want to add another model without percent2:
regress csat percent high
Now type to export the results to excel (notice we add the append option)
outreg2 using results, bdec(2) tdec(2) rdec(2) adec(2) alpha(0.001, 0.01, 0.05) addstat(Adj. R-
squared, e(r2_a)) excel append
In excel In word