Introduction to

Descriptive
Statistics
17.871
Key measures
Describing data

Moment Non-mean based

measure
Center Mean Mode, median

Spread Variance Range,

(standard deviation) Interquartile range

Skew Skewness --

Peaked Kurtosis --
Key distinction
Population vs. Sample Notation

Population vs. Sample

Greeks Romans
μ, σ, β s, b
Mean
n

 i
x
i 1
X
n
Variance, Standard Deviation
n
( xi   )2


i 1 n
 ,
2

n
( xi   )
2


i 1 n

 Variance, S.D. of a Sample
n
( xi   )2


i 1 n 1
s ,
2

Degrees of freedom
n
( xi   )
2


i 1 n 1
s
Binary data

X  prob( X )  1  proportion of time x  1

s  x (1  x )  s x  x (1  x )
2
x
 Normal distribution example
 IQ
Frequency  SAT
 Height

 “No skew”
 “Zero skew”
 Symmetrical
Value
 Mean = median = mode
1 ( x   ) / 2 2
f ( x)  e
 2
 Skewness
Asymmetrical distribution
Frequency  Income
 Contribution to
candidates
 Populations of
countries
 “Residual vote” rates

Value  “Positive skew”

 “Right skew”
 Skewness
Asymmetrical distribution
Frequency
 GPA of MIT students

 “Negative skew”
 “Left skew”

Value
Skewness
Frequency

Value
Kurtosis
k>3 leptokurtic
Frequency

k=3 mesokurtic

k<3 platykurtic

Value
Normal distribution
 Skewness = 0
 Kurtosis = 3

1 ( x   ) / 2 2
f ( x)  e
 2
More words about the normal curve
The z-score
or the
“standardized score”

z x x
x
Commands in STATA for
univariate statistics
 summarize varname
 summarize varname, detail
 histogram varname, bin() start() width()
density/fraction/frequency normal
 graph box varnames
 tabulate [NB: compare to table]
Example of Sophomore Test
Scores
 High School and Beyond, 1980: A Longitudinal
Survey of Students in the United States (ICPSR
Study 7896)

 totalscore = % of questions answered correctly

minus penalty for guessing
 recodedtype = (1=public school, 2=religious
private, 3 = non-sectarian private)
Explore totalscore some more

. table recodedtype,c(mean totalscore)

--------------------------
recodedty |
pe | mean(totals~e)
----------+---------------
1 | .3729735
2 | .4475548
3 | .589883
--------------------------
Graph totalscore
. hist totalscore

2
1.5
Density

1
.5
0

-.5 0 .5 1
totalscore
 Divide into “bins” so that each bar
represents 1% correct
 hist totalscore,width(.01)
 (bin=124, start=-.24209334, width=.01)

2
1.5
Density

1
.5
0

-.5 0 .5 1
totalscore
 Add ticks at each 10% mark
histogram totalscore, width(.01) xlabel(-.2 (.1) 1)
(bin=124, start=-.24209334, width=.01)
2
1.5
Density

1
.5
0

-.2 -.1 0 .1 .2 .3 .4 .5 .6 .7 .8 .9 1
totalscore
 Superimpose the normal curve
(with the same mean and s.d. as the empirical distribution)

. histogram totalscore, width(.01) xlabel(-.2 (.1) 1)

normal
(bin=124, start=-.24209334, width=.01)
2
1.5
Density

1
.5
0

-.2 -.1 0 .1 .2 .3 .4 .5 .6 .7 .8 .9 1
totalscore
 Histograms by category
.histogram totalscore, width(.01) xlabel(-.2 (.1)1)
by(recodedtype)
(bin=124, start=-.24209334, width=.01)

1 2
3

Public Religious private

2
1
0
Density

-.2 -.1 0 .1 .2 .3 .4 .5 .6 .7 .8 .9 1

3
3

Nonsectarian private
2
1
0

-.2 -.1 0 .1 .2 .3 .4 .5 .6 .7 .8 .9 1
totalscore
Graphs by recodedtype
Main issues with histograms
 Proper level of aggregation
 Non-regular data categories
 A note about histograms with
unnatural categories
From the Current Population Survey (2000), Voter and Registration Survey

How long (have you/has name) lived at this address?

-9 No Response
-3 Refused
-2 Don't know
-1 Not in universe
1 Less than 1 month
2 1-6 months
3 7-11 months
4 1-2 years
5 3-4 years
6 5 years or longer
 Solution, Step 1
Map artificial category onto
“natural” midpoint
-9 No Response  missing
-3 Refused  missing
-2 Don't know  missing
-1 Not in universe  missing
1 Less than 1 month  1/24 = 0.042
2 1-6 months  3.5/12 = 0.29
3 7-11 months  9/12 = 0.75
4 1-2 years  1.5
5 3-4 years  3.5
6 5 years or longer  10 (arbitrary)
Graph of recoded data
histogram longevity, fraction

.557134
Fraction

0
0 1 2 3 4 5 6 7 8 9 10
longevity
Density plot of data
Total area of last bar = .557
Width of bar = 11 (arbitrary)
Solve for: a = w h (or)
.557 = 11h => h = .051

0
0 1 2 3 4 5 6 7 8 9 10 15
longevity
Density plot template
Height
Category Fraction X-min X-max X-length (density)
< 1 mo. .0156 0 1/12 .082 .19*

1-6 mo. .0909 1/12 ½ .417 .22

7-11 mo. .0430 ½ 1 .500 .09

1-2 yr. .1529 1 2 1 .15

3-4 yr. .1404 2 4 2 .07

5+ yr. .5571 4 15 11 .05

* = .0156/.082
 Draw the previous graph with a box
plot
. graph box totalscore
1

Upper quartile
Inter-quartile
.5

Median } range
Lower quartile

} 1.5 x IQR
0
-.5
 Draw the box plots for the different
types of schools
. graph box totalscore, by(recodedtype)

1 2
1
.5
0
-.5

3
1
.5
0
-.5

Graphs by recodedtype
Draw the box plots for the different
types of schools using “over” option
graph box totalscore, over(recodedtype)
1
.5
0
-.5

1 2 3
Three words about pie charts:
don’t use them
So, what’s wrong with them
 For non-time series data, hard to get a
comparison among groups; the eye is very
bad in judging relative size of circle slices
 For time series, data, hard to grasp cross-
time comparisons
Some words about graphical
presentation
 Aspects of graphical integrity (following
Edward Tufte, Visual Display of
Quantitative Information)
 Main point should be readily apparent
 Show as much data as possible
 Write clear labels on the graph
 Show data variation, not design variation