© 2002-2008 The Trustees of Indiana University Univariate Analysis and Normality Test: 1
http://www.indiana.edu/~statmath
I n d i a n a U n i v e r s i t y
Universi t y I nf ormation T echnology Services
Univariate Analysis and Normality Test Using SAS, Stata,
and SPSS*
Hun Myoung Park, Ph.D.
© 2002-2008
Last modified on November 2008
University Information Technology Services
Center for Statistical and Mathematical Computing
Indiana University
410 North Park Avenue Bloomington, IN 47408
(812) 855-4724 (317) 278-4740
http://www.indiana.edu/~statmath
*
The citation of this document should read: “Park, Hun Myoung. 2008. Univariate Analysis and Normality Test
Using SAS, Stata, and SPSS. Working Paper. The University Information Technology Services (UITS) Center for
Statistical and Mathematical Computing, Indiana University.”
http://www.indiana.edu/~statmath/stat/all/normality/index.html
© 2002-2008 The Trustees of Indiana University Univariate Analysis and Normality Test: 2
http://www.indiana.edu/~statmath
This document summarizes graphical and numerical methods for univariate analysis and
normality test, and illustrates how to do using SAS 9.1, Stata 10 special edition, and SPSS 16.0.
1. Introduction
2. Graphical Methods
3. Numerical Methods
4. Testing Normality Using SAS
5. Testing Normality Using Stata
6. Testing Normality Using SPSS
7. Conclusion
1. Introduction
Descriptive statistics provide important information about variables to be analyzed. Mean,
median, and mode measure central tendency of a variable. Measures of dispersion include
variance, standard deviation, range, and interquantile range (IQR). Researchers may draw a
histogram, stem-and-leaf plot, or box plot to see how a variable is distributed.
Statistical methods are based on various underlying assumptions. One common assumption is
that a random variable is normally distributed. In many statistical analyses, normality is often
conveniently assumed without any empirical evidence or test. But normality is critical in many
statistical methods. When this assumption is violated, interpretation and inference may not be
reliable or valid.
Figure 1. Comparing the Standard Normal and a Bimodal Probability Distributions
0.1.2.3.4
-5 -3 -1 1 3 5
Standard Normal Distribution
0.1.2.3.4
-5 -3 -1 1 3 5
Bimodal Distribution
The t-test and ANOVA (Analysis of Variance) compare group means, assuming a variable of
interest follows a normal probability distribution. Otherwise, these methods do not make much
sense. Figure 1 illustrates the standard normal probability distribution and a bimodal
distribution. How can you compare means of these two random variables?
There are two ways of testing normality (Table 1). Graphical methods visualize the
distributions of random variables or differences between an empirical distribution and a
theoretical distribution (e.g., the standard normal distribution). Numerical methods present
© 2002-2008 The Trustees of Indiana University Univariate Analysis and Normality Test: 3
http://www.indiana.edu/~statmath
summary statistics such as skewness and kurtosis, or conduct statistical tests of normality.
Graphical methods are intuitive and easy to interpret, while numerical methods provide
objective ways of examining normality.
Table 1. Graphical Methods versus Numerical Methods
Graphical Methods Numerical Methods
Descriptive Stem-and-leaf plot, (skeletal) box plot,
dot plot, histogram
Skewness
Kurtosis
Theory-driven P-P plot
Q-Q plot
Shapiro-Wilk, Shapiro- Francia test
Kolmogorov-Smirnov test (Lillefors test)
Anderson-Darling/Cramer-von Mises tests
Jarque-Bera test, Skewness-Kurtosis test
Graphical and numerical methods are either descriptive or theory-driven. A dot plot and
histogram, for instance, are descriptive graphical methods, while skewness and kurtosis are
descriptive numerical methods. The P-P and Q-Q plots are theory-driven graphical methods for
normality test, whereas the Shapiro-Wilk W and Jarque-Bera tests are theory-driven numerical
methods.
Figure 2. Histograms of Normally and Non-normally Distributed Variables
0.1.2.3.4.5
-3 -2 -1 0 1 2 3
Randomly Drawn from the Standard Normal Distribution (Seed=1,234,567)
A Normally Distributed Variable (N=500)
0.03.06.09.12.15
0 10 20 30 40 50 60
Per Capita Gross National Income in 2005 ($1,000)
A Non-normally Distributed Variable (N=164)
Three variables are employed here. The first variable is unemployment rate of Illinois, Indiana,
and Ohio in 2005. The second variable includes 500 observations that were randomly drawn
from the standard normal distribution. This variable is supposed to be normally distributed with
mean 0 and variance 1 (left plot in Figure 2). An example of a non-normal distribution is per
capita gross national income (GNI) in 2005 of 164 countries in the world. GNIP is severely
skewed to the right and is least likely to be normally distributed (right plot in Figure 2). See the
Appendix for details.
© 2002-2008 The Trustees of Indiana University Univariate Analysis and Normality Test: 4
http://www.indiana.edu/~statmath
2. Graphical Methods
Graphical methods visualize the distribution of a random variable and compare the distribution
to a theoretical one using plots. These methods are either descriptive or theory-driven. The
former method is based on the empirical data, whereas the latter considers both empirical and
theoretical distributions.
2.1 Descriptive Plots
Among frequently used descriptive plots are the stem-and-leaf-plot, dot plot, (skeletal) box plot,
and histogram. When N is small, a stem-and-leaf plot and dot plot are useful to summarize
continuous or event count data. Figure 3 and 4 respectively present a stem-and-leaf plot and a
dot plot of the unemployment rate of three states.
Figure 3. Stem-and-Leaf Plot of Unemployment Rate of Illinois, Indiana, Ohio
-> state = IL
Stem-and-leaf plot for rate(Rate)
rate rounded to nearest multiple
of .1
plot in units of .1
3. | 7889
4* | 011122344
4. | 556666666677778888999
5* | 0011122222333333344444
5. | 5555667777777888999
6* | 000011222333444
6. | 555579
7* | 0033
7. |
8* | 0
8. | 8
-> state = IN
Stem-and-leaf plot for rate(Rate)
rate rounded to nearest multiple
of .1
plot in units of .1
3* | 1
3. | 89
4* | 012234
4. | 566666778889999
5* | 00000111222222233344
5. | 555666666777889
6* | 002222233344
6. | 5666677889
7* | 1113344
7. | 67
8* | 14
-> state = OH
Stem-and-leaf plot for rate (Rate)
rate rounded to nearest multiple
of .1
plot in units of .1
3* | 8
4* | 014577899
5* | 01223333445556667778888888999
6* | 001111122222233444446678899
7* | 01223335677
8* | 1223338
9* | 99
10* | 1
11* |
12* |
13* | 3
Figure 4. Dot Plot of Unemployment Rate of Illinois, Indiana, Ohio
051015
Illinois (N=102) Indiana (N=92) Ohio (N=88)
Indiana Business Research Center (http://www.stats.indiana.edu/)
Source: Bureau of Labor Statistics
© 2002-2008 The Trustees of Indiana University Univariate Analysis and Normality Test: 5
http://www.indiana.edu/~statmath
A box plot presents the minimum, 25th
percentile (1st
quartile), 50th
percentile (median), 75th
percentile (3rd
quartile), and maximum in a box and lines.1
Outliers, if any, appear at the
outsides of (adjacent) minimum and maximum lines. As such, a box plot effectively
summarizes these major percentiles using a box and lines. If a variable is normally distributed,
its 25th
and 75th
percentile are symmetric, and its median and mean are located at the same
point exactly in the center of the box.2
In Figure 5, you should see outliers in Illinois and Ohio that affect the shapes of corresponding
boxes. By contrast, the Indiana unemployment rate does not have outliers, and its symmetric
box implies that the rate appears to be normally distributed.
Figure 5. Box Plots of Unemployment Rates of Illinois, Indiana, and Ohio
2468101214
Illinois (N=102) Indiana (N=92) Ohio (N=88)
UnemploymentRate(%)
Indiana Business Research Center (http://www.stats.indiana.edu/)
Source: Bureau of Labor Statistics
The histogram graphically shows how each category (interval) accounts for the proportion of
total observations and is appropriate when N is large (Figure 6).
Figure 6. Histograms of Unemployment Rates of Illinois, Indiana and Ohio
0.1.2.3.4.5
0 3 6 9 12 15 0 3 6 9 12 15 0 3 6 9 12 15
Illinois (N=102) Indiana (N=92) Ohio (N=88)
Indiana Business Research Center (http://www.stats.indiana.edu/)
Source: Bureau of Labor Statistics
1
The first quartile cuts off lowest 25 percent of data; the second quartile, median, cuts data set in half; and the
third quartile cuts off lowest 75 percent or highest 25 percent of data. See http://en.wikipedia.org/wiki/Quartile
2
SAS reports a mean as “+” between (adjacent) minimum and maximum lines.
© 2002-2008 The Trustees of Indiana University Univariate Analysis and Normality Test: 6
http://www.indiana.edu/~statmath
2.2 Theory-driven Plots
P-P and Q-Q plots are considered here. The probability-probability plot (P-P plot or percent
plot) compares an empirical cumulative distribution function of a variable with a specific
theoretical cumulative distribution function (e.g., the standard normal distribution function). In
Figure 7, Ohio appears to deviate more from the fitted line than Indiana.
Figure 7. P-P Plots of Unemployment Rates of Indiana and Ohio (Year 2005)
0.000.250.500.751.00
0.00 0.25 0.50 0.75 1.00
Empirical P[i] = i/(N+1)
Source: Bureau of Labor Statistics
2005 Indiana Unemployment Rate (N=92 Counties)
0.000.250.500.751.00
0.00 0.25 0.50 0.75 1.00
Empirical P[i] = i/(N+1)
Source: Bureau of Labor Statistics
2005 Ohio Unemployment Rate (N=88 Counties)
Similarly, the quantile-quantile plot (Q-Q plot) compares ordered values of a variable with
quantiles of a specific theoretical distribution (i.e., the normal distribution). If two distributions
match, the points on the plot will form a linear pattern passing through the origin with a unit
slope. P-P and Q-Q plots are used to see how well a theoretical distribution models the
empirical data. In Figure 8, Indiana appears to have a smaller variation in its unemployment
rate than Ohio. By contrast, Ohio appears to have a wider range of outliers in the upper extreme.
Figure 8. Q-Q Plots of Unemployment Rates of Indiana and Ohio (Year 2005)
4.15.57.4
051015
UnemploymentRatein2005
5.641304 7.3501913.932418
3 4 5 6 7 8
Inverse Normal
Grid lines are 5, 10, 25, 50, 75, 90, and 95 percentiles
Source: Bureau of Labor Statistics
2005 Indiana Unemployment Rate (N=92 Counties)
4.56.18.8
051015
UnemploymentRatein2005
6.3625 8.7608573.964143
2 4 6 8 10
Inverse Normal
Grid lines are 5, 10, 25, 50, 75, 90, and 95 percentiles
Source: Bureau of Labor Statistics
2005 Ohio Unemployment Rate (N=88 Counties)
Detrended normal P-P and Q-Q plots depict the actual deviations of data points from the
straight horizontal line at zero. No specific pattern in a detrended plot indicates normality of the
variable. SPSS can generate detrended P-P and Q-Q plots.
© 2002-2008 The Trustees of Indiana University Univariate Analysis and Normality Test: 7
http://www.indiana.edu/~statmath
3. Numerical Methods
Graphical methods, although visually appealing, do not provide objective criteria to determine
normality of variables. Interpretations are thus a matter of judgments. Numerical methods use
descriptive statistics and statistical tests to examine normality.
3.1 Descriptive Statistics
Measures of dispersion such as variance reveal how observations of a random variable deviate
from their mean. The second central moment is
1
)( 2
2


 
n
xx
s i
Skewness is a third standardized moment that measures the degree of symmetry of a probability
distribution. If skewness is greater than zero, the distribution is skewed to the right, having
more observations on the left.
232
3
3
3
3
3
])([
)(1
)1(
)(])[(









xx
xxn
ns
xxxE
i
ii


Kurtosis, based on the fourth central moment, measures the thinness of tails or “peakedness” of
a probability distribution.
22
4
4
4
4
4
])([
)()1(
)1(
)(])[(









xx
xxn
ns
xxxE
i
ii


Figure 9. Probability Distributions with Different Kurtosis
0.2.4.6.80.2.4.6.8
-5 -3 -1 1 3 5
-5 -3 -1 1 3 5
Kurtosis < 3 Kurtosis = 3
Kurtosis > 3
© 2002-2008 The Trustees of Indiana University Univariate Analysis and Normality Test: 8
http://www.indiana.edu/~statmath
If kurtosis of a random variable is less than three (or if kurtosis-3 is less than zero), the
distribution has thicker tails and a lower peak compared to a normal distribution (first plot in
Figure 9).3
By contrast, kurtosis larger than 3 indicates a higher peak and thin tails (last plot). A
normally distributed random variable should have skewness and kurtosis near zero and three,
respectively (second plot in Figure 9).
state | N mean median max min variance skewness kurtosis
-------+--------------------------------------------------------------------------------
IL | 102 5.421569 5.35 8.8 3.7 .8541837 .6570033 3.946029
IN | 92 5.641304 5.5 8.4 3.1 1.079374 .3416314 2.785585
OH | 88 6.3625 6.1 13.3 3.8 2.126049 1.665322 8.043097
-------+--------------------------------------------------------------------------------
Total | 282 5.786879 5.65 13.3 3.1 1.473955 1.44809 8.383285
----------------------------------------------------------------------------------------
In short, skewness and kurtosis show how the distribution of a variable deviates from a normal
distribution. These statistics are based on the empirical data.
3.2 Theory-driven Statistics
The numerical methods of normality test include the Kolmogorov-Smirnov (K-S) D test
(Lilliefors test), Shapiro-Wilk test, Anderson-Darling test, and Cramer-von Mises test (SAS
Institute 1995).4
The K-S D test and Shapiro-Wilk W test are commonly used. The K-S,
Anderson-Darling, and Cramer-von Misers tests are based on the empirical distribution
function (EDF), which is defined as a set of N independent observations x1, x2, …xn with a
common distribution function F(x) (SAS 2004).
Table 2. Numerical Methods of Testing Normality
Test Statistic N Range Dist. SAS Stata SPSS
Jarque-Bera 2
 )2(2
 - - Skewness-Kurtosis
2
 9≤N
)2(2
 - .sktest Shapiro-Wilk
W 7≤N≤ 2,000 - YES .swilk YES
Shapiro-Francia W’ 5≤N≤ 5,000 - - .sfrancia Kolmogorov-Smirnov
D EDF YES *
YES
Cramer-vol Mises W2
EDF YES - Anderson-Darling
A2
EDF YES - *
Stata .ksmirnov command is not used for testing normality.
The Shapiro-Wilk W is the ratio of the best estimator of the variance to the usual corrected sum
of squares estimator of the variance (Shapiro and Wilk 1965).5
The statistic is positive and less
than or equal to one. Being close to one indicates normality.
3
SAS and SPSS produce (kurtosis -3), while Stata returns the kurtosis. SAS uses its weighted kurtosis formula
with the degree of freedom adjusted. So, if N is small, SAS, Stata, and SPSS may report different kurtosis.
4
The UNIVARIATE and CAPABILITY procedures have the NORMAL option to produce four statistics.
5
The W statistic was constructed by considering the regression of ordered sample values on corresponding
expected normal order statistics, which for a sample from a normally distributed population is linear (Royston
1982). Shapiro and Wilk’s (1965) original W statistic is valid for the sample sizes between 3 and 50, but Royston
extended the test by developing a transformation of the null distribution of W to approximate normality throughout
the range between 7 and 2000.
© 2002-2008 The Trustees of Indiana University Univariate Analysis and Normality Test: 9
http://www.indiana.edu/~statmath
The W statistic requires that the sample size is greater than or equal to 7 and less than or equal
to 2,000 (Shapiro and Wilk 1965).6
 



 2
2
)(
)( xx
xa
W
i
ii
where a’=(a1, a2, …, an) = 21111
]'[' 
mVVmVm , m’=(m1, m2, …, mn) is the vector of expected
values of standard normal order statistics, V is the n by n covariance matrix, x’=(x1, x2, …, xn)
is a random sample, and x(1)< x(2)< …<x(n).
The Shapiro-Francia W’ test is an approximate test that modifies the Shapro-Wilk W. The S-F
statistic uses b’=(b1, b2, …, bn) = 21
)'(' 
mmm instead of a’. The statistic was developed by
Shapiro and Francia (1972) and Royston (1983). The recommended sample sizes for the
Stata .sfrancia command range from 5 to 5,000 (Stata 2005). SAS and SPSS do not support
this statistic. Table 3 summarizes test statistics for 2005 unemployment rates of Illinois, Indiana,
and Ohio. Since N is not large, you need to read Shapiro-Wilk, Shapiro-Francia, Jarque-Bera,
and Skewness-Kurtosis statistics.
Table 3. Normality Test for 2005 Unemployment Rates of Illinois, Indiana, and Ohio
State Illinois Indiana Ohio
Test P-value Test P-value Test P-value
Shapiro-Wilk sas
.9714 .0260 .9841 .3266 .8858 .0001
Shapiro-Wilk stata
.9728 .0336 .9855 .4005 .8869 .0000
Shapiro-Francia stata
.9719 .0292 .9858 .3545 .8787 .0000
Kolmogorov-Smirnov sas
.0583 .1500 .0919 .0539 .1602 .0100
Cramer-von Misers sas
.0606 .2500 .1217 .0582 .4104 .0050
Anderson-Darling sas
.4534 .2500 .6332 .0969 2.2815 .0050
Jarque-Bera 12.2928 .0021 1.9458 .3380 149.5495 .0000
Skewness-Kurtosis stata
10.59 .0050 1.99 .3705 43.75 .0000
The SAS UNIVARIATE and CAPABILITY procedures perform the Kolmogorov-Smirnov D,
Anderson-Darling A2
, and Cramer-von Misers W2
tests, which are useful especially when N is
larger than 2,000.
3.3 Jarque-Bera (Skewness-Kurtosis) Test
The test statistics mentioned in the previous section tend to reject the null hypothesis when N
becomes large. Given a large number of observations, the Jarque-Bera test and SkewnessKurtosis
test will be alternative ways of normality test.
The Jarque-Bera test, a type of Lagrange multiplier test, was developed to test normality,
heteroscedasticy, and serial correlation (autocorrelation) of regression residuals (Jarque and
Bera 1980). The Jarque-Bera statistic is computed from skewness and kurtosis and
asymptotically follows the chi-squared distribution with two degrees of freedom.
6
Stata .swilk command, based on Shapiro and Wilk (1965) and Royston (1992), can be used with from 4 to
2000 observations (Stata 2005).
© 2002-2008 The Trustees of Indiana University Univariate Analysis and Normality Test: 10
http://www.indiana.edu/~statmath
)2(~
24
)3(
6
2
22





 

kurtosisskewness
n , where n is the number of observations.
The above formula gives a penalty for increasing the number of observations and thus implies a
good asymptotic property of the Jarque-Bera test. The computation for 2005 unemployment
rates is as follows.7
For Illinois: 12.292825 = 102*(0.66685022^2/6 + 1.0553068^2/24)
For Indiana: 1.9458304 = 92*(0.34732004^2/6 + (-0.1583764)^2/24)
For Ohio: 149.54945 = 88*(1.69434105^2/6 + 5.4132289^2/24)
The Stata Skewness-Kurtosis test is based on D’Agostino, Belanger, and D’Agostino, Jr. (1990)
and Royston (1991) (Stata 2005). Note that in Ohio the Jarque-Bera statistic of 150 is quite
different from the S-K statistic of 44 (see Table 3).
Table 4 Comparison of Methods for Testing Normality
N 10 100 500 1,000 5,000 10,000
Mean .5240 -.0711 -.0951 -.0097 -.0153 -.0192
Standard deviation .9554 1.0701 1.0033 1.0090 1.0107 1.0065
Minimum -.8659 -2.8374 -2.8374 -2.8374 -3.5387 -3.9838
1st
quantile -.2372 -.8674 -.8052 -.7099 -.7034 -.7121
Median .6411 -.0625 -.1196 -.0309 -.0224 -.0219
3rd
quantile 1.4673 .7507 .6125 .7027 .6623 .6479
Maximum 1.7739 1.9620 2.5117 3.1631 3.5498 4.3140
Skewness sas -.1620 -.2272 -.0204 .0100 .0388 .0391
Kurtosis-3 sas -1.4559 -.5133 -.3988 -.2633 -.0067 -.0203
Jarque-Bera .9269
(.6291)
1.9580
(.3757)
3.3483
(.1875)
2.9051
(.2340)
1.2618
(.5321)
2.7171
(.2570)
Skewness stata -.1366 -.2238 -.0203 .0100 .0388 .0391
Kurtosis stata 1.6310 2.4526 2.5932 2.7320 2.9921 2.9791
S-K stata 1.52
(.4030)
2.52
(.2843)
4.93
(.0850)
3.64
(.1620)
1.26
(.5330)
2.70
(.2589)
Shapiro-Wilk W sas .9359
(.5087)
.9840
(.2666)
.9956
(.1680)
.9980
(.2797)
.9998
(.8727)
.9999
(.8049)
Shapiro-F W’stata .9591
(.7256)
.9873
(.3877)
.9965
(.2941)
.9983
(.4009)
.9998
(.1000)
.9998
(.1000)
Kolmogorov-S Dsas .1382
(.1500)
.0708
(.1500)
.0269
(.1500)
.0180
(.1500)
.0076
(.1500)
.0073
(.1500)
Cramer-M W2 sas .0348
(.2500)
.0793
(.2167)
.0834
(.1945)
.0607
(.2500)
.0304
(.2500)
.0652
(.2500)
Anderson-D A2 sas .2526
(.2500)
.4695
(.2466)
.5409
(.1712)
.4313
(.2500)
.1920
(.2500)
.4020
(.2500)
* P-value in parentheses
Table 4 presents results of normality tests for random variables with different numbers of
observations. The data were randomly generated from the standard normal distribution with a
seed of 1,234,567 in SAS. As N grows, the mean, median, skewness, and (kurtosis-3) approach
zero, and the standard deviation gets close to 1. The Kolmogorov-Smirnov D, Anderson-
7
Skewness and Kurtosis are computed using the SAS UNIVARIATE and CAPABILITY procedures that report
kurtosis minus 3.
© 2002-2008 The Trustees of Indiana University Univariate Analysis and Normality Test: 11
http://www.indiana.edu/~statmath
Darling A2
, Cramer-von Mises W2
are computed in SAS, while the Skewness-Kurtosis and
Shapiro-Francia W’ are computed in Stata.
All four statistics do not reject the null hypothesis of normality regardless of the number of
observations (Table 4). Note that the Shapiro-Wilk W is not reliable when N is larger than
2,000 and S-F W’ is valid up to 5,000 observations. The Jarque-Bera and Skewness-Kurtosis
tests show consistent results.
3.4 Software Issues
The UNIVARIATE procedure of SAS/BASE and CAPABILITY of SAS/QC compute various
statistics and produce P-P and Q-Q plots. These procedures provide many numerical methods
including Cramer-vol Mises and Anderson-Darling.8
The P-P plot is generated only in
CAPABILITY.
By contrast, Stata has many individual commands to examine normality. In particular, Stata
provides .sktest and .sfrancia to conduct Skewness-Kurtosis and Shapiro-Francia W’ tests,
respectively.
SPSS EXAMINE provides numerical and graphical methods for normality test. The detrended
P-P and Q-Q plots can be generated in SPSS. Since SPSS has changed graph-related features
over time, you need to check menus, syntaxes, and reported bugs.
Table 5 summarizes SAS procedures and Stata/SPSS commands that are used to test normality
of random variables.
Table 5. Comparison of Procedures and Commands Available
SAS Stata SPSS
Descriptive statistics
(Skewness/Kurtosis)
UNIVARIATE .summarize
.tabstat
Descriptives, Frequencies
Examine
Histogram, dot plot UNIVARIATE
CHART, PLOT
.histogram
.dotplot
Graph, Igraph, Examine,
Frequencies
Stem-leaf-plot UNIVARIATE*
.stem Examine
Box plot UNIVARIATE*
.graph box Examine, Igraph
P-P plot CAPABILITY**
.pnorm Pplot
Q-Q plot UNIVARIATE .qnorm Pplot, Examine
Detrended Q-Q/P-P plot Pplot, Examine
Jarque-Bera (S-K) test .sktest
Shapiro-Wilk W UNIVARIATE .swilk Examine
Shapiro-Francia W’ .sfrancia
Kolmogorov-Smirnov UNIVARIATE Examine
Cramer-vol Mises UNIVARIATE
Anderson-Darling UNIVARIATE
* The UNIVARIATE procedure can provide the plot.
** The CAPABILITY procedure can provide the plot.
8
MINITAB also performs the Kolmogorov-Smirnov and Anderson-Darling tests.
© 2002-2008 The Trustees of Indiana University Univariate Analysis and Normality Test: 12
http://www.indiana.edu/~statmath
4. Testing Normality in SAS
SAS has the UNIVARIATE and CAPABILITY procedures to compute descriptive statistics,
draw various graphs, and conduct statistical tests for normality. Two procedures have similar
usage and produce similar statistics in the same format. However, UNIVARIATE produces a
stem-and-leaf plot, box plot, and normal probability plot, while CAPABILITY provides P-P
plot and CDP plot that UNIVARIATE does not.
This section illustrates how to summarize normally and non-normally distributed variables and
conduct normality tests of these variables using the two procedures (see Figure 10).
Figure 10. Histogram of Normally and Non-normally Distributed Variables
Cur ve: Nor mal ( Mu=- 0. 095 Si gma=1. 0033)
P
e
r
c
e
n
t
0
5
10
15
20
25
r andom
- 3. 5 - 3. 0 - 2. 5 - 2. 0 - 1. 5 - 1. 0 - 0. 5 0. 0 0. 5 1. 0 1. 5 2. 0 2. 5
Cur ve: Nor mal ( Mu=8. 9646 Si gma=13. 567)
P
e
r
c
e
n
t
0
10
20
30
40
50
60
GNI P
0 8 16 24 32 40 48 56 64
4.1 A Normally Distributed Variable
The UNIVARIATE procedure provides a variety of descriptive statistics, Q-Q plot, leaf-andstem-plot,
and box plot. This procedure also conducts Kolmogorov-Smirnov test, ShapiroWilk’
test, Anderson-Darling, and Cramer-von Misers tests.
Let us take a look at an example of the UNIVARIATE procedure. The NORMAL option
conducts normality testing; PLOT draws a leaf-and-stem plot and a box plot; finally, the
QQPLOT statement draws a Q-Q plot.
PROC UNIVARIATE DATA=masil.normality NORMAL PLOT;
VAR random;
QQPLOT random /NORMAL(MU=EST SIGMA=EST COLOR=RED L=1);
RUN;
Like UNIVARIATE, the CAPABILITY procedure also produces various descriptive statistics
and plots. CAPABILITY can draw a P-P plot using the PPPLOT option but does not support a
leaf-and-stem plot, a box plot, and a normal probability plot; this procedure does not have the
PLOT option available in UNIVARIATE.
4.1.1 SAS Output of Descriptive Statistics
© 2002-2008 The Trustees of Indiana University Univariate Analysis and Normality Test: 13
http://www.indiana.edu/~statmath
The following is an example of the CAPABILITY procedure. QQPLOT, PPPLOT, and
HISTOGRAM statements respectively draw a Q-Q plot, P-P plot, and histogram. Note that the
INSET statement adds summary statistics to graphs such as histogram and Q-Q plot.
PROC CAPABILITY DATA=masil.normality NORMAL;
VAR random;
QQPLOT random /NORMAL(MU=EST SIGMA=EST COLOR=RED L=1);
PPPLOT random /NORMAL(MU=EST SIGMA=EST COLOR=RED L=1);
HISTOGRAM /NORMAL(COLOR=MAROON W=4) CFILL = BLUE CFRAME = LIGR;
INSET MEAN STD /CFILL=BLANK FORMAT=5.2 ;
RUN;
The CAPABILITY Procedure
Variable: random
Moments
N 500 Sum Weights 500
Mean -0.0950725 Sum Observations -47.536241
Std Deviation 1.00330171 Variance 1.00661432
Skewness -0.0203721 Kurtosis -0.3988198
Uncorrected SS 506.819932 Corrected SS 502.300544
Coeff Variation -1055.3019 Std Error Mean 0.04486902
Basic Statistical Measures
Location Variability
Mean -0.09507 Std Deviation 1.00330
Median -0.11959 Variance 1.00661
Mode . Range 5.34911
Interquartile Range 1.41773
Tests for Location: Mu0=0
Test -Statistic- -----p Value------
Student's t t -2.11889 Pr > |t| 0.0346
Sign M -28 Pr >= |M| 0.0138
Signed Rank S -6523 Pr >= |S| 0.0435
Tests for Normality
Test --Statistic--- -----p Value-----
Shapiro-Wilk W 0.995564 Pr < W 0.168
Kolmogorov-Smirnov D 0.026891 Pr > D >0.150
Cramer-von Mises W-Sq 0.083351 Pr > W-Sq 0.195
Anderson-Darling A-Sq 0.540894 Pr > A-Sq 0.171
Quantiles (Definition 5)
© 2002-2008 The Trustees of Indiana University Univariate Analysis and Normality Test: 14
http://www.indiana.edu/~statmath
Quantile Estimate
100% Max 2.511694336
99% 2.055464409
95% 1.530450397
90% 1.215210586
75% Q3 0.612538495
50% Median -0.119592165
25% Q1 -0.805191028
10% -1.413548051
5% -1.794057126
1% -2.219479314
0% Min -2.837417522
Extreme Observations
-------Lowest------- -------Highest------
Value Obs Value Obs
-2.83741752 29 2.14897641 119
-2.59039285 204 2.21109349 340
-2.47829639 73 2.42113892 325
-2.39126554 391 2.42171307 139
-2.24047386 393 2.51169434 332
4.1.2 Graphical Methods
The stem-and-leaf plot and box plot, produced by the UNIVARIATE produre, illustrate that the
variable is normally distributed (Figure 11). The locations of first quantile, mean, median, and
third quintile indicate a bell-shaped distribution. Note that the mean -.0951 and median -.1196
are very close.
Figure 11. Stem-and-Leaf Plot and Box Plot of a Normally Distributed Variable
Histogram # Boxplot
2.75+* 1 |
.** 4 |
.******** 23 |
.**************** 46 |
.*********************** 68 +-----+
.*************************** 80 | |
.*************************************** 116 *--+--*
.********************** 64 +-----+
.******************* 56 |
.********* 27 |
.***** 13 |
-2.75+* 2 |
----+----+----+----+----+----+----+----
* may represent up to 3 counts
The normal probability plot available in UNIVARIATE shows a straight line, implying the
normality of the randomly drawn variable (Figure 12).
© 2002-2008 The Trustees of Indiana University Univariate Analysis and Normality Test: 15
http://www.indiana.edu/~statmath
Figure 12. Normal Probability Plot of a Normally Distributed Variable
Normal Probability Plot
2.75+ *
| +++**
| ********
| *******
| *****
| ******
| *******
| *****
| ******
| ******
|*******
-2.75+*+
+----+----+----+----+----+----+----+----+----+----+
-2 -1 0 +1 +2
The P-P and Q-Q plots below show that the data points are not seriously deviated from the
fitted line. They consistently indicate that the variable is normally distributed.
Figure 13. P-P plot and Q-Q Plot of a Normally Distributed Variable
C
u
m
u
l
a
t
i
v
e
D
i
s
t
r
i
b
u
t
i
o
n
o
f
r
a
n
d
o
m
0. 0
0. 2
0. 4
0. 6
0. 8
1. 0
Nor mal ( Mu=- 0. 095 Si gma=1. 0033)
0. 0 0. 2 0. 4 0. 6 0. 8 1. 0 - 4 - 3 - 2 - 1 0 1 2 3 4
- 3
- 2
- 1
0
1
2
3
r
a
n
d
o
m
Nor mal Quant i l es
4.1.3 Numerical Methods
The mean of -.0951 is very close to 0 and variance is almost 1. The skewness and kurtosis-3 are
respectively -.0204 and -.3988, indicating an almost normal distribution. However, these
descriptive statistics do not provide conclusive information about normality.
SAS provides four different statistics for testing normality. Shapiro-Wilk W of .9956 does not
reject the null hypothesis that the variable is normally distributed (p<.168). Similarly,
Kolmogorov-Smirnov, Cramer-von Mises, and Anderson-Darling tests do not reject the null
hypothesis. Since the number of observations is less than 2,000, however, Shapiro-Wilk W test
will be appropriate for this case.
© 2002-2008 The Trustees of Indiana University Univariate Analysis and Normality Test: 16
http://www.indiana.edu/~statmath
The Jarque-Bera test also indicates the normality of the randomly drawn variable (p=.1875).
Note that -.3988 is kurtosis -3.
)2(3.3482776~
24
0.3988198-
6
0.0203721-
500
22







Consequently, we can safely conclude that the randomly drawn variable is normally distributed.
4.2 A Non-normally Distributed Variable
Let us examine the per capita gross national income as an example of non-normally distributed
variables. See the appendix for details about this variable.
4.2.1 SAS Output of Descriptive Statistics
This section employs the UNIVARIATE procedure to compute descriptive statistics and
perform normality tests. The variable has mean 8.9646 and median 2.0495, where are
substantially different. Variance 184.0577 is extremely large.
PROC UNIVARIATE DATA=masil.gnip NORMAL PLOT;
VAR gnip;
QQPLOT gnip /NORMAL(MU=EST SIGMA=EST COLOR=RED L=1);
HISTOGRAM / NORMAL(COLOR=MAROON W=4) CFILL = BLUE CFRAME = LIGR;
RUN;
The UNIVARIATE Procedure
Variable: GNIP
Moments
N 164 Sum Weights 164
Mean 8.9645732 Sum Observations 1470.19001
Std Deviation 13.5667877 Variance 184.057728
Skewness 2.04947469 Kurtosis 3.60816725
Uncorrected SS 43181.0356 Corrected SS 30001.4096
Coeff Variation 151.337798 Std Error Mean 1.05938813
Basic Statistical Measures
Location Variability
Mean 8.964573 Std Deviation 13.56679
Median 2.765000 Variance 184.05773
Mode 1.010000 Range 65.34000
Interquartile Range 7.72500
Tests for Location: Mu0=0
Test -Statistic- -----p Value------
© 2002-2008 The Trustees of Indiana University Univariate Analysis and Normality Test: 17
http://www.indiana.edu/~statmath
Student's t t 8.462029 Pr > |t| <.0001
Sign M 82 Pr >= |M| <.0001
Signed Rank S 6765 Pr >= |S| <.0001
Tests for Normality
Test --Statistic--- -----p Value------
Shapiro-Wilk W 0.663114 Pr < W <0.0001
Kolmogorov-Smirnov D 0.284426 Pr > D <0.0100
Cramer-von Mises W-Sq 4.346966 Pr > W-Sq <0.0050
Anderson-Darling A-Sq 22.23115 Pr > A-Sq <0.0050
Quantiles (Definition 5)
Quantile Estimate
100% Max 65.630
99% 59.590
95% 38.980
90% 32.600
75% Q3 8.680
50% Median 2.765
25% Q1 0.955
10% 0.450
5% 0.370
1% 0.290
0% Min 0.290
Extreme Observations
----Lowest---- ----Highest----
Value Obs Value Obs
0.29 164 46.32 5
0.29 163 47.39 4
0.31 162 54.93 3
0.33 161 59.59 2
0.34 160 65.63 1
4.2.2 Graphical Methods
The stem-and-leaf plot, box plot, and normal probability plots all indicate that the variable is
not normally distributed (Figure 14). Most observations are highly concentrated on the left side
of the distribution. See the stem-and-leaf plot and box plot in Figure 14.
Figure 14. Stem-and-Leaf Plot, Box Plot, and Normally Probability Plot
Histogram # Boxplot
67.5+* 1 *
© 2002-2008 The Trustees of Indiana University Univariate Analysis and Normality Test: 18
http://www.indiana.edu/~statmath
.
.* 1 *
.* 1 *
.* 2 *
.* 3 *
.** 6 *
.** 5 0
.** 5 0
.* 2 0
.** 6 |
.*** 7 |
.****** 17 +--+--+
2.5+************************************ 108 *-----*
----+----+----+----+----+----+----+*
may represent up to 3 counts
Normal Probability Plot
67.5+ *
|
| *
| *
| **
| ** +++
| *** +++
| ** ++++
| **+++
| +*+
| ++++**
| ++++ **
| +++ ****
2.5+* * *** **********************
+----+----+----+----+----+----+----+----+----+----+
-2 -1 0 +1 +2
The following P-P and Q-Q plots show that the data points are seriously deviated from the
fitted line (Figure 15).
Figure 15. P-P plot and Q-Q Plot of a Non-normally Distributed Variable
C
u
m
u
l
a
t
i
v
e
D
i
s
t
r
i
b
u
t
i
o
n
o
f
G
N
I
P
0. 0
0. 2
0. 4
0. 6
0. 8
1. 0
Normal ( Mu=8. 9646 Si gma=13. 567)
0. 0 0. 2 0. 4 0. 6 0. 8 1. 0 - 3 - 2 - 1 0 1 2 3
0
10
20
30
40
50
60
70
G
N
I
P
Nor mal Quant i l es
4.2.3 Numerical Methods
© 2002-2008 The Trustees of Indiana University Univariate Analysis and Normality Test: 19
http://www.indiana.edu/~statmath
Per capita gross national income has a mean of 8.9646 and a large variance of 184.0557. Its
skewness and kurtosis-3 are 2.0495 and 3.6082, respectively, indicating that the variable is
highly skewed to the right with a high peak and thin tails.
It is not surprising that the Shapiro-Wilk test rejected the null hypothesis; W is .6631 and pvalue
is less than .0001. Kolmogorov-Smirnov, Cramer-von Mises, and Anderson-Darling tests
also report similar results.
Finally, the Jarque-Bera test returns 203.7717, which rejects the null hypothesis of normality at
the .05 level (p<.0000).
)2(203.77176~
24
3.60816725
6
2.04947469
164
22







To sum, we can conclude that the per capita gross national income is not normally distributed.
© 2002-2008 The Trustees of Indiana University Univariate Analysis and Normality Test: 20
http://www.indiana.edu/~statmath
5. Testing Normality Using Stata
In Stata, you have to use individual commands to get specific statistics or draw various plots.
This section contrasts normally distributed and non-normally distributed variables using
graphical and numerical methods.
5.1 Graphical Methods
A histogram is the most widely used graphical method. The histograms of normally and nonnormally
distributed variables are presented in the introduction. The Stata .histogram
command is followed by a variable name and options. The normal option adds a normal
density curve to the histogram.
. histogram normal, normal
. histogram gnip, normal
Let us draw a stem-and-leaf plot using the .stem command. The stem-and-leaf plot of the
randomly drawn normal shows a bell-shaped distribution (Figure 16).
. stem normal
Figure 16. Stem-and-Leaf Plot of a Normally Distributed Variable
Stem-and-leaf plot for normal
normal rounded to nearest multiple of .01
plot in units of .01
-28* | 4
-27* |
-26* |
-25* | 9
-24* | 8
-23* | 9
-22* | 40
-21* | 93221
-20* | 8650
-19* | 8842
-18* | 875200
-17* | 94
-16* | 9987550
-15* | 97643320
-14* | 87755432110
-13* | 98777655433210
-12* | 8866666433210
-11* | 987774332210
-10* | 875322
-9* | 88887665542210
-8* | 99988777533110
-7* | 77766544100
-6* | 998332
-5* | 99988877654433221110
-4* | 9998766655444433321
-3* | 88766654433322221100
-2* | 999988766555544433322111100
-1* | 8888777776655544433222221110
-0* | 99887776655433333111
0* | 01233344445669
1* | 0111222333445666778
2* | 0001234444556889999
3* | 1133444556667899
© 2002-2008 The Trustees of Indiana University Univariate Analysis and Normality Test: 21
http://www.indiana.edu/~statmath
4* | 014455667777
5* | 00112334556888
6* | 0001123668899
7* | 00233466799999
8* | 1122334667889
9* | 012445666778889
10* | 1133457799
11* | 1222334445689
12* | 122233489
13* | 26889
14* | 2777799
15* | 00112459
16* | 1347
17* | 02467
18* | 358
19* | 03556
20* |
21* | 5
22* | 1
23* |
24* | 22
25* | 1
By contrast, per capita gross national income is highly skewed to the right, having most
observations within $10,000 (Figure 17).
. stem gnip
Figure 17. Stem-and-Leaf Plot of a Non-normally Distributed Variable
Stem-and-leaf plot for gnip
gnip rounded to nearest multiple of .1
plot in units of .1
0** | 03,03,03,03,03,03,03,03,04,04,04,04,04,04,04,04,04,04,05,05, ... (64)
0** | 21,22,23,23,23,24,24,24,24,25,25,25,26,26,26,27,28,28,28,28, ... (34)
0** | 44,45,45,46,46,47,48,48,50,50,50,52,53,55,59
0** | 62,68,71,71,73,76,79
0** | 81,82,83,91,91
1** | 00,04,07,09,18
1** | 36
1** | 44,58
1** | 62,65,74
1** | 86,97
2** |
2** | 38
2** | 40,54
2** | 60,75,77,78
2** |
3** | 00
3** | 22,26
3** | 46,48,57
3** | 66,70,75,76
3** | 90
4** | 02,11
4** | 37
4** |
4** | 63,74
4** |
5** |
5** |
5** | 49
5** |
5** | 96
6** |
6** |
6** | 56
© 2002-2008 The Trustees of Indiana University Univariate Analysis and Normality Test: 22
http://www.indiana.edu/~statmath
The .dotplot command generates a dot plot, very similar to the stem-and leaf plot, in a
descending order (Figure 18).
. dotplot normal
. dotplot gnip
Figure 18. Dotplots of Normally and Non-normally Distributed Variables
-3-2-1012
0 10 20 30 40
Frequency
A Normally Distributed Variable (N=500)
020406080
0 10 20 30 40
Frequency
A Non-normally Distributed Variable (N=164)
The .graph box command draws a box plot. In the left plot of Figure 19, the shaded box
represents the 25th
percentile, median, and 75th
percentile, which are symmetrically arranged.
The right plot has an asymmetric box with many outliers beyond the adjacent maximum line.
. graph box normal
. graph box gnip
Figure 19. Box plots of Normally and Non-normally Distributed Variables
-4-202
A Normally Distributed Variable (N=500)
020406080
A Non-normally Distributed Variable (N=164)
The .pnorm command produces standardized normal P-P plot. The left plot shows almost no
deviation from the line, while the right depicts an s-shaped curve that is largely deviated from
the fitted line. In Stata, a P-P plot has the cumulative distribution of an empirical variable on
the x axis and the theoretical normal distribution on the y axis.9
.pnorm normal
.pnorm gnip
9
In SAS, these distributions are located reversely.
© 2002-2008 The Trustees of Indiana University Univariate Analysis and Normality Test: 23
http://www.indiana.edu/~statmath
Figure 20. P-P plots of Normally and Non-normally Distributed Variables0.000.250.500.751.00
NormalF[(normal-m)/s]
0.00 0.25 0.50 0.75 1.00
Empirical P[i] = i/(N+1)
A Normally Distributed Variable (N=500)
0.000.250.500.751.00
NormalF[(gnip-m)/s]
0.00 0.25 0.50 0.75 1.00
Empirical P[i] = i/(N+1)
A Non-normally Distributed Variable (N=164)
The .qnorm command produces a standardized normal Q-Q plot. The following Q-Q plots
show a similar pattern that P-P plots do (Figure 21). In the right plot, data points are
systematically deviated from the straight fitted line.
.qnorm normal
.qnorm gnip
Figure 21. Q-Q plots of Normally and Non-normally Distributed Variables
-1.794057-.11959221.53045
-4-2024
-.0950725 1.555212-1.745357
-4 -2 0 2 4
Inverse Normal
Grid lines are 5, 10, 25, 50, 75, 90, and 95 percentiles
A Normally Distributed Variable (N=500)
.372.76538.98
-200204060
8.964573 31.27995-13.35081
-20 0 20 40
Inverse Normal
Grid lines are 5, 10, 25, 50, 75, 90, and 95 percentiles
A Non-normally Distributed Variable (N=164)
5.2 Numerical Methods
Let us first get summary statistics using the .summarize command. The detail option lists
various statistics including mean, standard deviation, minimum, and maximum. Skewness and
kurtosis of a randomly drawn variable are respectively close to 0 and 3, implying normality.
Per capital gross national income has large skewness of 2.03 and kurtosis of 6.46, being skewed
to the right with a high peak and flat tails.
. summarize normal, detail
normal
-------------------------------------------------------------
Percentiles Smallest
1% -2.219479 -2.837418
5% -1.794057 -2.590393
© 2002-2008 The Trustees of Indiana University Univariate Analysis and Normality Test: 24
http://www.indiana.edu/~statmath
10% -1.413548 -2.478296 Obs 500
25% -.805191 -2.391266 Sum of Wgt. 500
50% -.1195922 Mean -.0950725
Largest Std. Dev. 1.003302
75% .6125385 2.211093
90% 1.215211 2.421139 Variance 1.006614
95% 1.53045 2.421713 Skewness -.0203109
99% 2.055464 2.511694 Kurtosis 2.593181
. sum gnip, detail
gnip
-------------------------------------------------------------
Percentiles Smallest
1% .29 .29
5% .37 .29
10% .45 .31 Obs 164
25% .955 .33 Sum of Wgt. 164
50% 2.765 Mean 8.964573
Largest Std. Dev. 13.56679
75% 8.68 47.39
90% 32.6 54.93 Variance 184.0577
95% 38.98 59.59 Skewness 2.030682
99% 59.59 65.63 Kurtosis 6.462734
The .tabstat command is vary useful to produce descriptive statistics in a table form. The
column(variable)option lists statistics vertically (in table rows). The command for the
variable normal is skipped.
. tabstat gnip, stats(n mean sum max min range sd var semean skewness kurtosis ///
median p1 p5 p10 p25 p50 p75 p90 p95 p99 iqr q) column(variable)
stats | normal
---------+----------
N | 500
mean | -.0950725
sum | -47.53624
max | 2.511694
min | -2.837418
range | 5.349112
sd | 1.003302
variance | 1.006614
se(mean) | .044869
skewness | -.0203109
kurtosis | 2.593181
p50 | -.1195922
p1 | -2.219479
p5 | -1.794057
p10 | -1.413548
p25 | -.805191
p50 | -.1195922
p75 | .6125385
p90 | 1.215211
p95 | 1.53045
p99 | 2.055464
iqr | 1.41773
p25 | -.805191
p50 | -.1195922
p75 | .6125385
--------------------
stats | gnip
---------+----------
N | 164
mean | 8.964573
sum | 1470.19
max | 65.63
min | .29
range | 65.34
sd | 13.56679
variance | 184.0577
se(mean) | 1.059388
skewness | 2.030682
kurtosis | 6.462734
p50 | 2.765
p1 | .29
p5 | .37
p10 | .45
p25 | .955
p50 | 2.765
p75 | 8.68
p90 | 32.6
p95 | 38.98
p99 | 59.59
iqr | 7.725
p25 | .955
p50 | 2.765
p75 | 8.68
--------------------
Now let us conduct statistical tests of normality. Stata provide three testing methods: ShapiroWilk
test, Shapiro-Francia test, and Skewness-Kurtosis test. The .swilk and .sfrancia
commands respectively conduct the Shapiro-Wilk and Shapiro-Francia tests. Both tests do not
© 2002-2008 The Trustees of Indiana University Univariate Analysis and Normality Test: 25
http://www.indiana.edu/~statmath
reject normality of the randomly drawn variable and reject normality of per capita gross
national income.
. swilk normal
Shapiro-Wilk W test for normal data
Variable | Obs W V z Prob>z
-------------+-------------------------------------------------
normal | 500 0.99556 1.492 0.962 0.16804
. sfrancia normal
Shapiro-Francia W' test for normal data
Variable | Obs W' V' z Prob>z
-------------+-------------------------------------------------
normal | 500 0.99645 1.273 0.541 0.29412
. swilk gnip
Shapiro-Wilk W test for normal data
Variable | Obs W V z Prob>z
-------------+-------------------------------------------------
gnip | 164 0.66322 42.309 8.530 0.00000
. sfrancia gnip
Shapiro-Francia W' test for normal data
Variable | Obs W' V' z Prob>z
-------------+-------------------------------------------------
gnip | 164 0.66365 45.790 7.413 0.00001
Stata’s .sktest command conducts the Skewness-Kurtosis test that is conceptually similar to
the Jarque-Bera test. The noadjust option suppresses the empirical adjustment made by
Royston (1991). The following S-K tests do not reject normality of a randomly drawn variable
at the .05 level but surprisingly reject the null hypothesis at the .1 level.
. sktest normal
Skewness/Kurtosis tests for Normality
------- joint ------
Variable | Pr(Skewness) Pr(Kurtosis) adj chi2(2) Prob>chi2
-------------+-------------------------------------------------------
normal | 0.851 0.027 4.93 0.0850
. sktest normal, noadjust
Skewness/Kurtosis tests for Normality
------- joint ------
Variable | Pr(Skewness) Pr(Kurtosis) chi2(2) Prob>chi2
-------------+-------------------------------------------------------
normal | 0.851 0.027 4.93 0.0850
Like the Shapiro-Wilk and Shapiro-Francia tests, both S-K tests below reject the null
hypothesis that per capita gross national income is normally distributed at the .01 significance
level.
. sktest gnip
Skewness/Kurtosis tests for Normality
------- joint ------
Variable | Pr(Skewness) Pr(Kurtosis) adj chi2(2) Prob>chi2
-------------+-------------------------------------------------------
gnip | 0.000 0.000 55.33 0.0000
© 2002-2008 The Trustees of Indiana University Univariate Analysis and Normality Test: 26
http://www.indiana.edu/~statmath
. sktest gnip, noadjust
Skewness/Kurtosis tests for Normality
------- joint ------
Variable | Pr(Skewness) Pr(Kurtosis) chi2(2) Prob>chi2
-------------+-------------------------------------------------------
gnip | 0.000 0.000 75.39 0.0000
The Jarque-Bera statistic of normal is 3.4823 = 500*(-.0203109^2/6+(2.593181-3)^2/24),
which is not large enough to reject the null hypothesis (p<.1753). The Jarque-Bera statistic of
the per capita gross national income is 194.6489 = 164*(2.030682^2/6+(6.462734-3)^2/24).
This large chi-squared rejects the null hypothesis (p<.0000). The Jarque-Bera test appears to be
more reliable than the Stata S-K test (see Table 4).
In conclusion, graphical methods and numerical methods provide sufficient evidence that per
capita gross national income is not normally distributed.
© 2002-2008 The Trustees of Indiana University Univariate Analysis and Normality Test: 27
http://www.indiana.edu/~statmath
6. Testing Normality Using SPSS
SPSS has the DESCRIPTIVES and FREQUENCIES commands to produce descriptive
statistics. DESCRIPTIVES is usually applied to continuous variables, but FREQUENCIES is
also able to produce various descriptive statistics in addition to frequency tables. The IGRAPH
command draws histogram and box plots. The PPLOT command produces (detrended) P-P and
Q-Q plots.
The EXAMINE command can produce both descriptive statistics and various plots, such as a
stem-leaf-plot, histogram, box plot, (detrended) P-P plot, and (detrended) Q-Q plot. EXAMINE
also performs the Kolmogorov-Smirnov and Shapiro-Wilk tests for normality.
6.1 A Normally Distributed Variable
DESCRIPTIVES summarizes interval or continuous variables and FREQUENCIES reports
frequency tables of discrete variables and summary statistics. The /STATISTICS subcommand
in both commands specify statistics to be produced.
The following DESCRIPTIVES command reports the number of observations, sum, mean,
variance, standard deviation of normal.10
The mean of -.10 and standard deviation 1 implies
that the variable is normally distributed.
DESCRIPTIVES VARIABLES=normal
/STATISTICS=MEAN SUM STDDEV VARIANCE.
Descriptive Statistics
N Sum Mean Std. Deviation Variance
normal 500 -47.54 -.0951 1.00330 1.007
Valid N
(listwise)
500
The following FREQUENCIES produces various statistics of normal, a frequency table, and a
histogram.11
Since normal is continuous, its frequency table is long and thus skipped here. The
/HISTOGRAM subcommand draws a histogram, which is the same as what the GRAPH
command in the next page produces.
FREQUENCIES VARIABLES=normal /NTILES= 4
/STATISTICS=STDDEV VARIANCE RANGE MINIMUM MAXIMUM SEMEAN MEAN MEDIAN MODE
SUM SKEWNESS SESKEW KURTOSIS SEKURT
/HISTOGRAM
/ORDER= ANALYSIS.
Statistics
10
In order to execute this command, open a syntax window , copy and paste the syntax into the window, and then
click Run menu. Alternatively, click Analysis Descriptive StatisticsDescriptives and provide a variable of
interest.
11
Click Analysis Descriptive StatisticsFrequencies and then specify statistics using the Statistics option.
© 2002-2008 The Trustees of Indiana University Univariate Analysis and Normality Test: 28
http://www.indiana.edu/~statmath
normal
Valid 500.000N
Missing .000
Mean -.095
Std. Error of Mean .045
Median -.120
Mode -2.837a
Std. Deviation 1.003
Variance 1.007
Skewness -.020
Std. Error of Skewness .109
Kurtosis -.399
Std. Error of Kurtosis .218
Range 5.349
Minimum -2.837
Maximum 2.512
Sum -47.536
25 -.807
50 -.120
Percentiles
75 .613
a. Multiple modes exist. The smallest value is shown
The variable has a mean -.10 and a unit variance. The median -.120 is very close to the mean.
The kurtosis-3 is -.399 and skewness is -.020.
6.1.1 Graphical Methods
Like the /HISTOGRAM subcommand of FREQUENCIES, the GRAPH command draws a
histogram of the variable normal (left plot in Figure 22).12
GRAPH /HISTOGRAM=normal.
The IGRAPH command can produce a similar histogram (right plot in Figure 22) but its syntax
appears to be messy.13
Two histograms report mean -.1 and standard deviation 1 on the right
top corner and suggest that the variable is normally distributed.
IGRAPH /VIEWNAME='Histogram'
/X1 = VAR(normal) TYPE = SCALE
/Y = $count /COORDINATE = VERTICAL
/X1LENGTH=3.0 /YLENGTH=3.0
/X2LENGTH=3.0
/CHARTLOOK='NONE'
/Histogram SHAPE = HISTOGRAM CURVE = OFF X1INTERVAL AUTO X1START = 0.
Figure 22. Histogram of a Normally Distributed Variable
12
Click GraphsLegacy DialogsHistogram.
13
Click GraphsLegacy DialogsInteractiveHistogram.
© 2002-2008 The Trustees of Indiana University Univariate Analysis and Normality Test: 29
http://www.indiana.edu/~statmath
The EXAMINE command can produce descriptive statistics as well as a stem-and-leaf plot and
a box plot (Figure 23 and 24).14
The /PLOT subcommand with STEMLEAF and BOXPLOT
draws two plots that is very similar to the histogram in Figure 22.
EXAMINE VARIABLES=normal
/PLOT BOXPLOT STEMLEAF
/COMPARE GROUP
/STATISTICS DESCRIPTIVES
/CINTERVAL 95
/MISSING LISTWISE
/NOTOTAL.
Figure 23. Stem-and-Leaf Plot of a Normally Distributed Variable
normal Stem-and-Leaf Plot
Frequency Stem & Leaf
2.00 -2 . &
13.00 -2 . 00111&
27.00 -1 . 555566678899
56.00 -1 . 000111111222222333333344444
64.00 -0 . 555555555666777778888888999999
116.00 -0 . 000000000011111111111111222222222222233333333334444444444
80.00 0 . 000000011111111122222222233333333444444
68.00 0 . 555555556666677777778888889999999
46.00 1 . 000001111112222334444
23.00 1 . 55566778899
4.00 2 . 4&
1.00 2 . &
Stem width: 1.00
Each leaf: 2 case(s)
& denotes fractional leaves.
Figure 24. Box Plot of a Normally Distributed Variable
14
Click AnalyzeDescriptive StatisticsExplore, and then include the variable you want to examine.
© 2002-2008 The Trustees of Indiana University Univariate Analysis and Normality Test: 30
http://www.indiana.edu/~statmath
The both extremes (i.e., minimum and maximum), the 25th
, 50th
, and 75th
percentiles are
symmetrically arranged in the box plot.
EXAMINE also produces a histogram and normal Q-Q plot and detrended normal Q-Q plot
using HISTOGRAM and NPPLOT option (Figure 25).15
NPPLOT conducts normality test and
draw the two Q-Q plots.
EXAMINE VARIABLES=normal
/PLOT HISTOGRAM NPPLOT
/COMPARE GROUP /STATISTICS DESCRIPTIVES
/CINTERVAL 95 /MISSING LISTWISE /NOTOTAL.
Figure 25. Q-Q and Detrended Q-Q Plots of a Normally Distributed Variable
15
In the Explore dialog box, choose Plots option and then check Normality plots with tests option.
© 2002-2008 The Trustees of Indiana University Univariate Analysis and Normality Test: 31
http://www.indiana.edu/~statmath
The PPLOT command produces P-P and Q-Q plots as well. 16
The /TYPE subcommand
chooses either P-P or Q-Q plot and /DIST specifies a probability distribution (e.g., the standard
normal distribution). The following PPLOT command draws normal P-P and detrended normal
P-P plots (Figure 26); the output of other descriptive statistics is skipped here.
PPLOT /VARIABLES=normal
/NOLOG /NOSTANDARDIZE
/TYPE=Q-Q /FRACTION=BLOM /TIES=MEAN /DIST=NORMAL.
Figure 26. P-P and Detrended P-P Plots of a Normally Distributed Variable
The following PPLOT command draws normal Q-Q and detrended normal Q-Q plots of the
variable (see Figure 25).
PPLOT /VARIABLES=normal
/NOLOG /NOSTANDARDIZE
/TYPE=Q-Q /FRACTION=BLOM /TIES=MEAN /DIST=NORMAL.
Both P-P and Q-Q plots show no significant deviation from the fitted line. As in Stata, the
normal Q-Q plot and detrended Q-Q plot has observed quantiles on the X axis and normal
quantiles on the Y axis.
6.1.2 Numerical Methods
EXAMINE has the /PLOT NPPLOT subcommand to test normality of a variable. This
command produces descriptive statistics (/STATISTICS DESCRIPTIVES), outliers
(EXTREME), draws a normal Q-Q plot (/PLOT NPPLOT), and performs the KolmogorovSmirnov
and Shapiro-Wilk tests.
EXAMINE VARIABLES=normal
/PLOT NPPLOT
/STATISTICS DESCRIPTIVES EXTREME
/CINTERVAL 95 /MISSING LISTWISE /NOTOTAL.
Case Processing Summary
16
In SPSS 16.0, you may not see P-P and Q-Q under the Graphs menu, which were available in previous versions.
© 2002-2008 The Trustees of Indiana University Univariate Analysis and Normality Test: 32
http://www.indiana.edu/~statmath
Cases
Valid Missing Total
N Percent N Percent N Percent
normal 500 100.0% 0 .0% 500 100.0%
Descriptives
Statistic Std. Error
Mean -.0951 .04487
Lower Bound -.183295% Confidence Interval
for Mean
Upper Bound -.0069
5% Trimmed Mean -.0933
Median -.1196
Variance 1.007
Std. Deviation 1.00330
Minimum -2.84
Maximum 2.51
Range 5.35
Interquartile Range 1.42
Skewness -.020 .109
normal
Kurtosis -.399 .218
Extreme Values
Case Number Value
1 332 2.51
2 139 2.42
3 325 2.42
4 340 2.21
Highest
5 119 2.15
1 29 -2.84
2 204 -2.59
3 73 -2.48
4 391 -2.39
Normal
Lowest
5 393 -2.24
Since N is less than 2,000, we have to read the Shapiro-Wilk statistic and do not reject the null
hypothesis of normality (p<.168). Like SAS, SPSS reports the same Kolmogorov-Smirnov
statistic of .027, but it provides an adjusted p-value of .200, a bit larger than the .150 that SAS
reports.
© 2002-2008 The Trustees of Indiana University Univariate Analysis and Normality Test: 33
http://www.indiana.edu/~statmath
Tests of Normality
Kolmogorov-Smirnov
a
Shapiro-Wilk
Statistic df Sig. Statistic df Sig.
Normal .027 500 .200
*
.996 500 .168
a. Lilliefors Significance Correction
*. This is a lower bound of the true significance.
6.2 A Non-normally Distributed Variable
Let us consider per capita national gross income that is not normally distributed.
6.2.1 Graphical Methods
The following EXAMINE command produce the histogram, stem-and-leaf plot, and box plot of
a non-normally distributed variable gnip. The stem-and-leaf plot is skipped here.
EXAMINE VARIABLES=gnip
/PLOT BOXPLOT STEMLEAF HISTOGRAM NPPLOT
/STATISTICS DESCRIPTIVES EXTREME
/CINTERVAL 95 /MISSING LISTWISE /NOTOTAL.
Figure 27 illustrates that the distribution is heavily skewed to the right and there exist many
outliers beyond the extreme line in the box plot (right plot). The median and the 25th
percentile
are close to each other.
Figure 27. Histogram and Box Plot a Non-normally Distributed Variable
Figure 28 presents the P-P and detrended P-P plots where data points are significantly deviated
from the straight fitted line.
PPLOT /VARIABLES=gnip
/NOLOG /NOSTANDARDIZE
/TYPE=P-P /FRACTION=BLOM
/TIES=MEAN
© 2002-2008 The Trustees of Indiana University Univariate Analysis and Normality Test: 34
http://www.indiana.edu/~statmath
/DIST=NORMAL.
Figure 28. P-P and Detrended P-P Plots of a Non-normally Distributed Variable
The Q-Q and detrended Q-Q plots also show a significant deviation from the fitted line (Figure
26).
PPLOT /VARIABLES=gnip
/NOLOG /NOSTANDARDIZE
/TYPE=Q-Q /FRACTION=BLOM
/TIES=MEAN
/DIST=NORMAL.
Figure 29. Q-Q and Detrended Q-Q Plots of a Non-normally Distributed Variable
6.2.2 Numerical Methods
The descriptive statistics of gnip indicates that the variable is not normally distributed. There is
a large gap between the mean of 8.9646 and the median of 2.7650. The skewness and kurtosis -
3 are 2.049 and 3.608, respectively. The variable appears severely skewed to the right with a
higher peak and flat tails. The following tables are the output of the above EXAMINE
command.
© 2002-2008 The Trustees of Indiana University Univariate Analysis and Normality Test: 35
http://www.indiana.edu/~statmath
Case Processing Summary
Cases
Valid Missing Total
N Percent N Percent N Percent
gnip 164 100.0% 0 .0% 164 100.0%
Descriptives
Statistic Std. Error
Mean 8.9646 1.05939
Lower Bound 6.872795% Confidence Interval
for Mean
Upper Bound 11.0565
5% Trimmed Mean 7.1877
Median 2.7650
Variance 184.058
Std. Deviation 13.56679
Minimum .29
Maximum 65.63
Range 65.34
Interquartile Range 7.92
Skewness 2.049 .190
gnip
Kurtosis 3.608 .377
Extreme Values
Case Number Value
1 1 65.63
2 2 59.59
3 3 54.93
4 4 47.39
Highest
5 5 46.32
1 164 .29
2 163 .29
3 162 .31
4 161 .33
gnip
Lowest
5 160 .34a
a. Only a partial list of cases with the value .34
are shown in the table of lower extremes.
Tests of Normality
© 2002-2008 The Trustees of Indiana University Univariate Analysis and Normality Test: 36
http://www.indiana.edu/~statmath
Kolmogorov-Smirnova
Shapiro-Wilk
Statistic df Sig. Statistic df Sig.
gnip .284 164 .000 .663 164 .000
a. Lilliefors Significance Correction
The Shapiro-Wilk test rejects the null hypothesis of normality at the .05 level. The Jarque-Bera
test also rejects the null hypothesis with a large statistic of 204. Its computation is skipped (see
section 4.2.3). Based on a consistent result from both graphical and numerical methods, we can
conclude the variable gnip is not normally distributed.
© 2002-2008 The Trustees of Indiana University Univariate Analysis and Normality Test: 37
http://www.indiana.edu/~statmath
7. Conclusion
Univariate analysis is the first step of data analysis once a data set is ready. Various descriptive
statistics provide valuable basic information about variables that is used to determine
appropriate analysis methods to be employed.
Normality is commonly assumed in many statistical and economic methods, although often
conveniently assumed in reality without any empirical test. Violation of this assumption will
result in unreliable inferences and misleading interpretations.
There are graphical and numerical methods for conducting univariate analysis and normality
tests (Table 1). Graphical methods produce various plots such as a stem-and-leaf plot,
histogram, and a P-P plot that are intuitive and easy to interpret. Some are descriptive and
others are theory-driven.
Numerical methods compute a variety of measures of central tendency and dispersion such as
mean, median, quantile, variance, and standard deviation. Skewness and kurtosis provide clues
to the normality of a variable. If skewness and kurtosis-3 are close to zero, the variable may be
normally distributed. Keep in mind that SAS and SPSS report kurtosis-3, while Stata returns
kurtosis itself.
If the skewness of a varialbe is larger than 0, the variable is skewed to the right with many
observations on the left of the distribution; a negative skewness indicates many observations on
the right. If kurtosis-3 is greater than 0 (or kurtosis is greater than 3), the distribution has a high
peak and flat tails (third plot in Figure 8). If kurtosis is smaller than 3, the variable has a low
peak and thick tails (first plot in Figure 9).
In addition to these descriptive statistics, there are formal ways to perform normality tests. The
Shapiro-Wilk and Shapiro-Francia tests are proper when N is less than 2,000 and 5,000,
respectively. The Kolmogorov-Smirnov, Cramer-vol Mises, and Anderson-Darling tests are
recommended when N is large. The Jarque-Bera test, although not supported by most statistical
software packages, is a consistent method of normality testing.
The SAS UNIVARIATE and CONTENTS procedures provide a variety of descriptive statistics
and normality testing methods including Kolmogorov-Smirnov, Cramer-vol Mises, and
Anderson-Darling tests (Table 5). These procedures produce stem-and-leaf, box plot, histogram,
P-P plot, and Q-Q plot as well. Stata has various commands for univariate analysis and
graphics. In particular, Stata supports the Shapiro-Francia test, a modification of the ShapiroWilk
test, and the skewness-kurtosis test. But there is no command to conduct the KolmogorovSmirnov
test for normality in Stata. SPSS can produce detrended P-P and Q-Q plots, and
perform the Shapiro-Wilk and Kolmogorov-Smirnov tests with Lilliefors significance
correction.
© 2002-2008 The Trustees of Indiana University Univariate Analysis and Normality Test: 38
http://www.indiana.edu/~statmath
Appendix A: Data Sets
This document uses the following three variables.
1. Unemployment Rate of Illinois, Indiana, and Ohio in 2005
This unemployment rate is provided by Bureau of Labor Statistics. Actual data were
downloaded from http://www.stats.indiana.edu/, Indiana Business Research Center of the
Kelley School of Business, Indiana University.
. tabstat rate, stat(mean sd p25 median p75 skewness kurtosis) by(state)
Summary for variables: rate
by categories of: state
state | mean sd p25 p50 p75 skewness kurtosis
-------+----------------------------------------------------------------------
IL | 5.421569 .9242206 4.7 5.35 6 .6570033 3.946029
IN | 5.641304 1.038929 4.9 5.5 6.35 .3416314 2.785585
OH | 6.3625 1.458098 5.5 6.1 6.95 1.665322 8.043097
-------+----------------------------------------------------------------------
Total | 5.786879 1.214066 5 5.65 6.4 1.44809 8.383285
------------------------------------------------------------------------------
2. A Randomly Drawn Variable
This variable includes 500 observations that were randomly drawn from the standard normal
distribution with a seed of 1,234,567. The RANNOR() of SAS was used as a random number
generator.
%LET n=500; %LET dataset=n500;
DATA masil.&dataset;
seed=1234567;
DO i=1 TO &n;
normal=RANNOR(seed); OUTPUT;
END;
RUN;
. tabstat normal, stat(mean sd p25 median p75 skewness kurtosis)
variable | mean sd p25 p50 p75 skewness kurtosis
-------------+----------------------------------------------------------------------
normal | -.0950725 1.003302 -.805191 -.1195922 .6125385 -.0203109 2.593181
------------------------------------------------------------------------------------
3. Per Capita Gross National Income in 2005.
This data set includes per capita gross national incomes of 164 countries in the world that are
provided by World Bank (http://web.worldbank.org/).
. tabstat gnip, stat(mean sd p25 median p75 skewness kurtosis)
variable | mean sd p25 p50 p75 skewness kurtosis
-------------+----------------------------------------------------------------------
gnip | 8.964573 13.56679 .955 2.765 8.68 2.030682 6.462734
------------------------------------------------------------------------------------
© 2002-2008 The Trustees of Indiana University Univariate Analysis and Normality Test: 39
http://www.indiana.edu/~statmath
© 2002-2008 The Trustees of Indiana University Univariate Analysis and Normality Test: 40
http://www.indiana.edu/~statmath
References
Bera, Anil. K., and Carlos. M. Jarque. 1981. "Efficient Tests for Normality, Homoscedasticity
and Serial Independence of Regression Residuals: Monte Carlo Evidence." Economics
Letters, 7(4):313-318.
D’Agostino, Ralph B., Albert Belanger, and Ralph B. D’Agostino, Jr. 1990. “A Suggestion for
Using Powerful and Informative Tests of Normality.” American Statistician, 44(4): 316-
321.
Jarque, Carlos M., and Anil K. Bera. 1980. "Efficient Tests for Normality, Homoscedasticity
and Serial Independence of Regression Residuals." Economics Letters, 6(3):255-259.
Jarque, Carlos M., and Anil K. Bera. 1987. "A Test for Normality of Observations and
Regression Residuals." International Statistical Review, 55(2):163-172.
Mitchell, Michael N. 2004. A Visual Guide to Stata Graphics. College Station, TX: Stata Press.
Royston, J. P. 1982. "An Extension of Shapiro and Wilk's W Test for Normality to Large
Samples." Applied Statistics, 31(2): 115-124.
Royston, J. P. 1983. "A Simple Method for Evaluating the Shapiro-Francia W' Test of NonNormality."
Statistician, 32(3) (September): 297-300.
Royston, J.P. 1991. “Comment on sg3.4 and an Improved D’Agostino test.” Stata Technical
Bulletin, 3: 13-24.
Royston, P.J. 1992. "Approximating the Shapiro-Wilk W-Test for Non-normality." Statistics
and Computing, 2:117-119.
SAS Institute. 1995. SAS/QC Software: Usage and Reference I and II. Cary, NC: SAS Institute.
SAS Institute. 2004. SAS 9.1.3 Procedures Guide Volume 4. Cary, NC: SAS Institute.
Shapiro, S. S., and M. B. Wilk. 1965. "An Analysis of Variance Test for Normality (Complete
Samples)." Biometrika, 52(3/4) (December):591-611.
Shapiro, S. S., and R. S. Francia. 1972. "An Approximate Analysis of Variance Test for
Normality." Journal of the American Statistical Association, 67 (337) (March): 215-216.
SPSS Inc. 2007. SPSS 16.0 Command Syntax Reference. Chicago, IL: SPSS Inc.
Stata Press. 2007. Stata Base Reference Manual Release 10. College Station, TX: Stata Press.
Stata Press. 2007. Stata Graphics Reference Manual Release 10. College Station, TX: Stata
Press.
© 2002-2008 The Trustees of Indiana University Univariate Analysis and Normality Test: 41
http://www.indiana.edu/~statmath
Acknowledgements
I am grateful to Jeremy Albright and Kevin Wilhite at the UITS Center for Statistical and
Mathematical Computing for comments and suggestions.
Revision History
 2002 First draft.
 2006. 11 Revision with new data.
 2008. 11 Revision with new versions of software packages.