Title stata.com
ci — Confidence intervals for means, proportions, and variances
Description Quick start Menu Syntax
Options Remarks and examples Stored results Methods and formulas
Acknowledgment References Also see
Description
ci computes confidence intervals for population means, proportions, variances, and standard
deviations.
cii is the immediate form of ci; see [U] 19 Immediate commands for a general discussion of
immediate commands.
Quick start
Confidence intervals for means of normally distributed variables v1, v2, and v3
ci means v1-v3
Confidence interval for mean of Poisson-distributed variable v4
ci means v4, poisson
Confidence interval for rate of v4 with total exposure recorded in v5
ci means v4, poisson exposure(v5)
Confidence interval for proportion of binary variable v6
ci proportions v6
Confidence intervals for variances of v1, v2, and v3
ci variances v1-v3
Same as above, but Bonett confidence intervals are produced
ci variances v1-v3, bonett
90% Bonett confidence intervals for standard deviations of v1, v2, and v3
ci variances v1-v3, sd bonett level(90)
Confidence interval for a mean based on a sample with 85 observations, a sample mean of 10, and a
standard deviation of 3
cii means 85 10 3
90% confidence interval for rate from a sample with 4,379 deaths over 11,394 person-years
cii means 11394 4379, poisson level(90)
Agresti–Coull confidence interval for proportion based on a sample with 2,377 observations and 136
successes
cii proportions 2377 136, agresti
1
2 ci — Confidence intervals for means, proportions, and variances
Bonett confidence interval for variance based on a sample with 20 observations, sample variance of 9,
and estimated kurtosis of 1.8
cii variances 20 9 1.8, bonett
Same as above, but with confidence interval for standard deviation
cii variances 20 3 1.8, sd bonett
Menu
ci
Statistics > Summaries, tables, and tests > Summary and descriptive statistics > Confidence intervals
cii for a normal mean
Statistics > Summaries, tables, and tests > Summary and descriptive statistics > Normal mean CI calculator
cii for a Poisson mean
Statistics > Summaries, tables, and tests > Summary and descriptive statistics > Poisson mean CI calculator
cii for a proportion
Statistics > Summaries, tables, and tests > Summary and descriptive statistics > Proportion CI calculator
cii for a variance
Statistics > Summaries, tables, and tests > Summary and descriptive statistics > Variance CI calculator
cii for a standard deviation
Statistics > Summaries, tables, and tests > Summary and descriptive statistics > Standard deviation CI calculator
ci — Confidence intervals for means, proportions, and variances 3
Syntax
Confidence intervals for means, normal distribution
ci mean s
varlist
if
in
weight
, options
cii mean s #
obs
#
mean
#
sd
, level(#)
Confidence intervals for means, Poisson distribution
ci mean s
varlist
if
in
weight
, poisson
exposure(varname) options
cii means #
exposure
#
events
, poisson
level(#)
Confidence intervals for proportions
ci prop ortions
varlist
if
in
weight
, prop options options
cii prop ortions #
obs
#
succ
, prop options level(#)
Confidence intervals for variances
ci var iances
varlist
if
in
weight
, bonett options
cii var iances #
obs
#
variance
, level(#)
cii var iances #
obs
#
variance
#
kurtosis
, bonett
level(#)
Confidence intervals for standard deviations
ci var iances
varlist
if
in
weight
, sd
bonett options
cii var iances #
obs
#
sd
, sd
level(#)
cii var iances #
obs
#
sd
#
kurtosis
, sd bonett
level(#)
#
obs
must be a positive integer. #
exposure
, #
sd
, and #
variance
must be a positive number. #
succ
and
#
events
must be a nonnegative integer or between 0 and 1. If the number is between 0 and 1,
Stata interprets it as the fraction of successes or events and converts it to an integer number
representing the number of successes or events. The computation then proceeds as if two integers
had been specified. If option bonett is specified, you must additionally specify #
kurtosis
with cii
variances.
4 ci — Confidence intervals for means, proportions, and variances
prop options Description
exact calculate exact confidence intervals; the default
wald calculate Wald confidence intervals
wilson calculate Wilson confidence intervals
agresti calculate Agresti–Coull confidence intervals
jeffreys calculate Jeffreys confidence intervals
options Description
level(#) set confidence level; default is level(95)
separator(#) draw separator line after every # variables; default is separator(5)
total add output for all groups combined (for use with by only)
by, collect, and statsby are allowed with ci, and collect is allowed with cii; see [U] 11.1.10 Prefix commands.
aweights are allowed with ci means for normal data, and fweights are allowed with all ci subcommands; see
[U] 11.1.6 weight.
Options
Options are presented under the following headings:
Options for ci and cii means
Options for ci and cii proportions
Options for ci and cii variances
Options for ci and cii means
Main
poisson specifies that the variables (or numbers for cii) are Poisson-distributed counts; exact Poisson
confidence intervals will be calculated. By default, confidence intervals for means are calculated
based on a normal distribution.
exposure(varname) is used only with poisson. You do not need to specify poisson if you specify
exposure(); poisson is assumed. varname contains the total exposure (typically a time or an
area) during which the number of events recorded in varlist was observed.
level(#) specifies the confidence level, as a percentage, for confidence intervals. The default is
level(95) or as set by set level; see [R] level.
separator(#) specifies how often separation lines should be inserted into the output. The default is
separator(5), meaning that a line is drawn after every five variables. separator(10) would
draw the line after every 10 variables. separator(0) suppresses the separation line.
total is used with the by prefix. It requests that in addition to output for each by-group, output be
added for all groups combined.
ci — Confidence intervals for means, proportions, and variances 5
Options for ci and cii proportions
Main
exact, wald, wilson, agresti, and jeffreys specify how binomial confidence intervals are to be
calculated.
exact is the default and specifies exact (also known in the literature as Clopper–Pearson [1934])
binomial confidence intervals.
wald specifies calculation of Wald confidence intervals.
wilson specifies calculation of Wilson confidence intervals.
agresti specifies calculation of Agresti–Coull confidence intervals.
jeffreys specifies calculation of Jeffreys confidence intervals.
See Brown, Cai, and DasGupta (2001) for a discussion and comparison of the different binomial
confidence intervals.
level(#) specifies the confidence level, as a percentage, for confidence intervals. The default is
level(95) or as set by set level; see [R] level.
separator(#) specifies how often separation lines should be inserted into the output. The default is
separator(5), meaning that a line is drawn after every five variables. separator(10) would
draw the line after every 10 variables. separator(0) suppresses the separation line.
total is used with the by prefix. It requests that in addition to output for each by-group, output be
added for all groups combined.
Options for ci and cii variances
Main
sd specifies that confidence intervals for standard deviations be calculated. The default is to compute
confidence intervals for variances.
bonett specifies that Bonett confidence intervals be calculated. The default is to compute normal-based
confidence intervals, which assume normality for the data.
level(#) specifies the confidence level, as a percentage, for confidence intervals. The default is
level(95) or as set by set level; see [R] level.
separator(#) specifies how often separation lines should be inserted into the output. The default is
separator(5), meaning that a line is drawn after every five variables. separator(10) would
draw the line after every 10 variables. separator(0) suppresses the separation line.
total is used with the by prefix. It requests that in addition to output for each by-group, output be
added for all groups combined.
Remarks and examples stata.com
Remarks are presented under the following headings:
Confidence intervals for means
Normal-based confidence intervals
Poisson confidence intervals
Confidence intervals for proportions
Confidence intervals for variances
Immediate form
6 ci — Confidence intervals for means, proportions, and variances
Confidence intervals for means
ci means computes a confidence interval for the population mean for each of the variables in
varlist.
Normal-based confidence intervals
Example 1: Normal-based confidence intervals
Without the poisson option, ci means produces normal-based confidence intervals that are correct
if the variable is normally distributed and asymptotically correct for all other distributions satisfying
the conditions of the central limit theorem.
. use https://www.stata-press.com/data/r18/auto
(1978 automobile data)
. ci means mpg price
Variable Obs Mean Std. err. [95% conf. interval]
mpg 74 21.2973 .6725511 19.9569 22.63769
price 74 6165.257 342.8719 5481.914 6848.6
The standard error of the mean of mpg is 0.67, and the 95% confidence interval is [ 19.96, 22.64 ].
We can obtain wider confidence intervals, 99%, by typing
. ci means mpg price, level(99)
Variable Obs Mean Std. err. [99% conf. interval]
mpg 74 21.2973 .6725511 19.51849 23.07611
price 74 6165.257 342.8719 5258.405 7072.108
Example 2: The by prefix
The by prefix breaks out the confidence intervals according to by-group; total adds an overall
summary. For instance,
. by foreign: ci means mpg, total
-> foreign = Domestic
Variable Obs Mean Std. err. [95% conf. interval]
mpg 52 19.82692 .657777 18.50638 21.14747
-> foreign = Foreign
Variable Obs Mean Std. err. [95% conf. interval]
mpg 22 24.77273 1.40951 21.84149 27.70396
-> Total
Variable Obs Mean Std. err. [95% conf. interval]
mpg 74 21.2973 .6725511 19.9569 22.63769
ci — Confidence intervals for means, proportions, and variances 7
Example 3: Controlling the format
You can control the formatting of the numbers in the output by specifying a display format for
the variable; see [U] 12.5 Formats: Controlling how data are displayed. For instance,
. format mpg %9.2f
. ci means mpg
Variable Obs Mean Std. err. [95% conf. interval]
mpg 74 21.30 0.67 19.96 22.64
Poisson confidence intervals
If you specify the poisson option, ci means assumes count data and computes exact Poisson
confidence intervals.
Example 4: Poisson confidence intervals
We have data on the number of bacterial colonies on a Petri dish. The dish has been divided into
36 small squares, and the number of colonies in each square has been counted. Each observation in
our dataset represents a square on the dish. The variable count records the number of colonies in
each square counted, which varies from 0 to 5.
. use https://www.stata-press.com/data/r18/petri, clear
. ci means count, poisson
Poisson exact
Variable Exposure Mean Std. err. [95% conf. interval]
count 36 2.333333 .2545875 1.861158 2.888825
ci reports that the average number of colonies per square is 2.33. If the expected number of colonies
per square were as low as 1.86, the probability of observing 2.33 or more colonies per square would
be 2.5%. If the expected number were as large as 2.89, the probability of observing 2.33 or fewer
colonies per square would be 2.5%.
Example 5: Option exposure()
The number of “observations”— how finely the Petri dish is divided—makes no difference. The
Poisson distribution is a function only of the count. In example 4, we observed a total of 2.33×36 = 84
colonies and a confidence interval of [ 1.86 ×36, 2.89 ×36 ] = [ 67, 104 ]. We would obtain the same
[ 67, 104 ] confidence interval if our dish were divided into, say, 49 squares rather than 36.
For the counts, it is not even important that all the squares be of the same size. For rates, however,
such differences do matter but in an easy-to-calculate way. Rates are obtained from counts by dividing
by exposure, which is typically a number multiplied by either time or an area. For our Petri dishes,
we divide by an area to obtain a rate, but if our example were cast in terms of being infected by a
disease, we might divide by person-years to obtain the rate. Rates are convenient because they are
easier to compare: we might have 2.3 colonies per square inch or 0.0005 infections per person-year.
So let’s assume that we wish to obtain the number of colonies per square inch and, moreover, that
not all the “squares” on our dish are of equal size. We have a variable called area that records the
area of each square:
8 ci — Confidence intervals for means, proportions, and variances
. ci means count, exposure(area)
Poisson exact
Variable Exposure Mean Std. err. [95% conf. interval]
count 3 28 3.055051 22.3339 34.66591
The rates are now in more familiar terms. In our sample, there are 28 colonies per square inch, and
the 95% confidence interval is [ 22.3, 34.7 ]. When we did not specify exposure(), ci means with
option poisson assumed that each observation contributed 1 to exposure.
Technical note
If there were no colonies on our dish, ci means with option poisson would calculate a one-sided
confidence interval:
. use https://www.stata-press.com/data/r18/petrinone
. ci means count, poisson
Poisson exact
Variable Exposure Mean Std. err. [95% conf. interval]
count 36 0 0 0 .1024689*
(*) one-sided, 97.5% confidence interval
Confidence intervals for proportions
The ci proportions command assumes binary (0/1) data and computes binomial confidence
intervals.
Example 6: Exact binomial (Clopper–Pearson) confidence interval
We have data on employees, including a variable marking whether the employee was promoted
last year.
. use https://www.stata-press.com/data/r18/promo
. ci proportions promoted
Binomial exact
Variable Obs Proportion Std. err. [95% conf. interval]
promoted 20 .1 .067082 .0123485 .3169827
The exact binomial, also known as the Clopper–Pearson (1934) interval, is computed by default.
Nominally, the interpretation of a 95% confidence interval is that under repeated samples or
experiments, 95% of the resultant intervals would contain the unknown parameter in question.
However, for binomial data, the actual coverage probability, regardless of method, usually differs
from that interpretation. This result occurs because of the discreteness of the binomial distribution,
which produces only a finite set of outcomes, meaning that coverage probabilities are subject to
discrete jumps and that the exact nominal level cannot always be achieved. Therefore, the term “exact
confidence interval” refers to its being derived from the binomial distribution, the distribution exactly
generating the data, rather than resulting in exactly the nominal coverage.
ci — Confidence intervals for means, proportions, and variances 9
For the Clopper–Pearson interval, the actual coverage probability is guaranteed to be greater
than or equal to the nominal confidence level, here 95%. Because of the way it is calculated—see
Methods and formulas—it may also be interpreted as follows: If the true probability of being promoted
were 0.012, the chances of observing a result as extreme or more extreme than the result observed
(20 × 0.1 = 2 or more promotions) would be 2.5%. If the true probability of being promoted were
0.317, the chances of observing a result as extreme or more extreme than the result observed (two
or fewer promotions) would be 2.5%.
Example 7: Other confidence intervals
The Clopper–Pearson interval is desirable because it guarantees nominal coverage; however, by
dropping this restriction, you may obtain accurate intervals that are not as conservative. In this vein,
you might opt for the Wilson (1927) interval,
. ci proportions promoted, wilson
Wilson
Variable Obs Proportion Std. err. [95% conf. interval]
promoted 20 .1 .067082 .0278665 .3010336
the Agresti–Coull (1998) interval,
. ci proportions promoted, agresti
Agresti--Coull
Variable Obs Proportion Std. err. [95% conf. interval]
promoted 20 .1 .067082 .0156562 .3132439
or the Bayesian-derived Jeffreys interval (Brown, Cai, and DasGupta 2001),
. ci proportions promoted, jeffreys
Jeffreys
Variable Obs Proportion Std. err. [95% conf. interval]
promoted 20 .1 .067082 .0213725 .2838533
Picking the best interval is a matter of balancing accuracy (coverage) against precision (average
interval length) and depends on sample size and success probability. Brown, Cai, and DasGupta (2001)
recommend the Wilson or Jeffreys interval for small sample sizes (≤40) yet favor the Agresti–Coull
interval for its simplicity, decent performance for sample sizes less than or equal to 40, and performance
comparable to Wilson or Jeffreys for sample sizes greater than 40. They also deem the Clopper–
Pearson interval to be “wastefully conservative and [. . .] not a good choice for practical use”, unless
of course one requires, at a minimum, the nominal coverage level.
Finally, the binomial Wald confidence interval is obtained by specifying the wald option. The
Wald interval is the one taught in most introductory statistics courses and, for the above, is simply,
for level 1 −α, Proportion±z
α/2
(Std. err.), where z
α/2
is the 1 −α/2 quantile of the standard
normal. Because its overall poor performance makes it impractical, the Wald interval is available
mainly for pedagogical purposes. The binomial Wald interval is also similar to the interval produced
by treating binary data as normal data and using ci means, with two exceptions. First, the calculation
of the standard error in ci proportions uses denominator n rather than n − 1, used for normal
data in ci means. Second, confidence intervals for normal data are based on the t distribution rather
than the standard normal. Of course, both discrepancies vanish as sample size increases.
10 ci — Confidence intervals for means, proportions, and variances
Technical note
Let’s repeat example 6, but this time with data in which there are no promotions over the observed
period:
. use https://www.stata-press.com/data/r18/promonone
. ci proportions promoted
Binomial exact
Variable Obs Proportion Std. err. [95% conf. interval]
promoted 20 0 0 0 .1684335*
(*) one-sided, 97.5% confidence interval
The confidence interval is [ 0, 0.168 ], and this is the confidence interval that most books publish. It
is not, however, a true 95% confidence interval because the lower tail has vanished. As Stata notes,
it is a one-sided, 97.5% confidence interval. If you wanted to put 5% in the right tail, you could type
ci proportions promoted, level(90).
Technical note
ci proportions ignores any variables that do not take on the values 0 and 1 exclusively. For
instance, with our automobile dataset,
. use https://www.stata-press.com/data/r18/auto
(1978 automobile data)
. ci proportions mpg foreign
Binomial exact
Variable Obs Proportion Std. err. [95% conf. interval]
foreign 74 .2972973 .0531331 .196584 .4148353
Note: The results are produced only for binary (0/1) variables.
We also requested the confidence interval for mpg, but Stata ignored us. It does that so you can type
ci proportions and obtain correct confidence intervals for all the variables that are 0/1 in your
data.
Confidence intervals for variances
The ci variances command computes confidence intervals for the variances or, if the sd option is
specified, for the standard deviations. The default is a normal-based confidence interval that assumes
the data are normal and uses a χ
2
distribution to construct the confidence intervals. If normality is
suspect, you may consider using the bonett option to compute Bonett (2006) confidence intervals,
which are more robust to nonnormality.
Example 8: Normal-based confidence intervals
So far, we have restricted our attention to confidence intervals for means and proportions. Typically,
when people think of statistical inference, they usually have in mind inferences concerning population
means. However, the population parameter of interest will vary from one situation to another. In many
scenarios, the population variance is as important as the population mean. For example, in a quality
ci — Confidence intervals for means, proportions, and variances 11
control study, a machine that fills 16-ounce canned peas is investigated at regular time intervals. A
random sample of n = 8 containers is selected every hour. Ideally, the amount of peas in a can should
vary only slightly about the 16-ounce value. If the variance was large, then a large proportion of
cans would be either underfilled, thus cheating the customer, or overfilled, thus resulting in economic
loss to the manufacturing company. Suppose that the weights of 16-ounce cans filled by the machine
are normally distributed. The acceptable variability in the weights is expected to be 0.09 with the
respective standard deviation of 0.3 ounces. To monitor the machine’s performance, we can compute
confidence intervals for the variance of the weights of cans:
. use https://www.stata-press.com/data/r18/peas_normdist
(Weights of canned peas, normal distribution)
. ci variances weight
Variable Obs Variance [95% conf. interval]
weight 8 .3888409 .1699823 1.610708
The command reports the sample estimate of the variance of 0.39 with the 95% confidence interval
of [ 0.17, 1.61 ].
Instead of the variance, we may be interested in confidence intervals for the standard deviation.
We can specify the sd option to compute such confidence intervals.
. ci variances weight, sd
Variable Obs Std. dev. [95% conf. interval]
weight 8 .6235711 .4122891 1.269137
The 95% confidence interval for the standard deviation of the weights is [0.41, 1.27]. Because the
desired value for the standard deviation, 0.3 ounces, falls outside the interval, the machine may require
some tuning.
Confidence intervals in example 8 are based on the assumption that the random sample is selected
from a population having a normal distribution. Nonnormality of the population distribution, in the
form of skewness or heavy tails, can have a drastic impact on the asymptotic coverage probability of
the normal-based confidence intervals. This is the case even for distributions that are similar to normal.
Scheff
´
e (1959, 336) showed that the normal-based interval has an asymptotic coverage probability
of about 0.76, 0.63, 0.60, and 0.51 for the logistic, t with seven degrees of freedom, Laplace, and t
with five degrees of freedom distributions, respectively. Miller (1997, 264) describes this situation as
“catastrophic” because these distributions are symmetric and not easily distinguishable from a normal
distribution unless the sample size is large. Hence, it is judicious to evaluate the normality of the
data prior to constructing the normal-based confidence intervals for variances or standard deviations.
Bonett (2006) proposed a confidence interval that performs well in small samples under moderate
departures from normality. His interval performs only slightly worse than the exact normal-based
confidence interval when sampling from a normal distribution. A larger sample size provides Bonett
confidence intervals with greater protection against nonnormality.
Example 9: Bonett confidence interval for normal data
We will repeat example 8 and construct a Bonett confidence interval for the standard deviation by
specifying the bonett option. The results are similar, and both examples lead to the same inferential
conclusion.
12 ci — Confidence intervals for means, proportions, and variances
. ci variances weight, sd bonett
Bonett
Variable Obs Std. dev. [95% conf. interval]
weight 8 .6235711 .3997041 1.288498
The Bonett confidence interval is wider than the normal-based confidence interval in example 8.
For normal data, Bonett (2006) suggested that if Bonett confidence interval is used for a sample of
size n + 3, then its average width will be about the same as the average width of the normal-based
confidence interval from a sample size of n. Sampling three more observations may be a small
price to pay because Bonett confidence intervals perform substantially better than the normal-based
confidence intervals for nonnormal data.
Example 10: Bonett confidence interval for nonnormal data
The following data have been generated from a t distribution with five degrees of freedom to
illustrate the effect of wrongfully using the normal-based confidence interval when the data-generating
process is not normal.
. use https://www.stata-press.com/data/r18/peas_tdist
(Weights of canned peas, t distribution)
. ci variances weight, sd
Variable Obs Std. dev. [95% conf. interval]
weight 8 2.226558 1.472143 4.531652
The standard deviation of a t distribution with five degrees of freedom is
p
5/3 ≈ 1.29 and falls
outside the confidence interval limits. If we suspect that data may not be normal, the Bonett confidence
interval is typically a better choice:
. ci variances weight, sd bonett
Bonett
Variable Obs Std. dev. [95% conf. interval]
weight 8 2.226558 1.137505 5.772519
The value 1.29 is within the limits of the Bonett confidence interval [ 1.14, 5.77 ]
Immediate form
So far, we computed confidence intervals for various parameters using data in memory. We can
also compute confidence intervals using only data summaries, without any data in memory. Each of
the considered ci commands has an immediate cii version that computes the respective confidence
intervals using data summaries.
Example 11: Confidence interval for a normal mean
We are reading a soon-to-be-published paper by a colleague. In it is a table showing the number
of observations, mean, and standard deviation of the 1980 median family income for the Northeast
and West. We correctly think that the paper would be much improved if it included the confidence
intervals. The paper claims that for 166 cities in the Northeast, the average of median family income
is $19,509 with a standard deviation of $4,379:
ci — Confidence intervals for means, proportions, and variances 13
For the Northeast:
. cii means 166 19509 4379
Variable Obs Mean Std. err. [95% conf. interval]
166 19509 339.8763 18837.93 20180.07
For the West:
. cii means 256 22557 5003
Variable Obs Mean Std. err. [95% conf. interval]
256 22557 312.6875 21941.22 23172.78
Example 12: Confidence interval for a Poisson mean
The number of reported traffic accidents in Santa Monica over a 24-hour period is 27. We need
know nothing else to compute a confidence interval for the mean number of accidents for a day:
. cii means 1 27, poisson
Poisson exact
Variable Exposure Mean Std. err. [95% conf. interval]
1 27 5.196152 17.79317 39.28358
Example 13: Confidence interval for a proportion
We flip a coin 10 times, and it comes up heads only once. We are shocked and decide to obtain
a 99% confidence interval for this coin:
. cii proportions 10 1, level(99)
Binomial exact
Variable Obs Proportion Std. err. [99% conf. interval]
10 .1 .0948683 .0005011 .5442871
Example 14: Confidence interval for a variance
A company fills 32-ounce tomato juice jars with a quantity of juice having a normal distribution
with a claimed variance not exceeding 0.2. A random sample of 15 jars is collected to evaluate this
claim. The sample variance is 0.5:
. cii variances 15 0.5
Variable Obs Variance [95% conf. interval]
15 .5 .2680047 1.243621
Because the advertised value of 0.2 does not fall inside the confidence interval, the company is
allowing too much variation in the amount of tomato juice per jar.
14 ci — Confidence intervals for means, proportions, and variances
Example 15: Confidence interval for a standard deviation
Suppose the director of statistical development at a statistical software company is a big soccer
fan and requires all developers to play on the company team in the city’s local soccer league. Ten
developers are randomly selected to participate in the game. To ensure an advantage over other
teams, the director requires each of the 10 developers to cover 6 miles on average each game. Being
merciful, she will tolerate a standard deviation of 0.3 miles across different players, arguing that this
will keep the team’s performance consistent. The distance covered by each player is measured using
a pedometer. At the end of the game, the sample standard deviation of the distances covered by the
10 players was 0.56 miles:
. cii variances 10 0.56, sd
Variable Obs Std. dev. [95% conf. interval]
10 .56 .3851877 1.022342
Because the confidence interval does not include the designated value for the standard deviation, 0.3
miles, it is clear the team is not meeting standards, and an unpleasant meeting is planned.
Example 16: Confidence interval for a standard deviation of nonnormal data
Continuing with example 15, a clever statistician points out that distances covered by company
players in a soccer match do not follow the normal distribution because some players, mostly
econometricians, walk on the field, while others, mostly statisticians, do all the running. Therefore,
the normal-based confidence interval (which assumes normality) is not valid. Instead, we should use
the Bonett confidence interval, which additionally requires an estimate of kurtosis; see Methods and
formulas. If kurtosis is estimated to be 5, we would obtain the following:
. cii variances 10 0.56 5, sd bonett
Bonett
Variable Obs Std. dev. [95% conf. interval]
10 .56 .2689449 1.45029
The Bonett confidence interval now contains the specified value for the standard deviation, 0.3 miles.
The director of statistics concludes that overall team performance is acceptable. An uncomfortable
meeting is still planned but for a smaller group.
Stored results
ci means and cii means store the following in r():
Scalars
r(N) number of observations or, if poisson is specified, exposure
r(mean) mean
r(se) estimate of standard error
r(lb) lower bound of confidence interval
r(ub) upper bound of confidence interval
r(level) confidence level of confidence interval
Macros
r(citype) normal or poisson; type of confidence interval
r(exposure) name of exposure variable with poisson
ci — Confidence intervals for means, proportions, and variances 15
ci proportions and cii proportions store the following in r():
Scalars
r(N) number of observations
r(proportion) proportion
r(se) estimate of standard error
r(lb) lower bound of confidence interval
r(ub) upper bound of confidence interval
r(level) confidence level of confidence interval
Macros
r(citype) exact, wald, wilson, agresti, or jeffreys; type of confidence interval
ci variances and cii variances store the following in r():
Scalars
r(N) number of observations
r(Var) variance
r(sd) standard deviation, if sd is specified
r(kurtosis) kurtosis, only if bonett is specified
r(lb) lower bound of confidence interval
r(ub) upper bound of confidence interval
r(level) confidence level of confidence interval
Macros
r(citype) normal or bonett, type of confidence interval
Methods and formulas
Methods and formulas are presented under the following headings:
Normal mean
Poisson mean
Binomial proportion
Variance and standard deviation
Normal mean
Define n, x, and s
2
as, respectively, the number of observations, (weighted) average, and (unbiased)
estimated variance of the variable in question; see [R] summarize.
The standard error of the mean, s
µ
, is defined as
p
s
2
/n.
Let α be 1 −l/100, where l is the confidence level specified by the user in the level() option.
Define t
α/2
as the two-sided t statistic corresponding to a significance level of α with n −1 degrees
of freedom; t
α/2
is obtained from Stata as invttail(n-1,0.5*α). The lower and upper confidence
bounds are, respectively, x − s
µ
t
α/2
and x + s
µ
t
α/2
.
Poisson mean
Given the total cases, k, the estimate of the expected count λ is k, and its standard error is
√
k.
ci means with option poisson calculates the exact confidence interval [ λ
1
, λ
2
] such that
Pr(K ≥ k|λ = λ
1
) = α/2
and
Pr(K ≤ k|λ = λ
2
) = α/2
where K is Poisson with mean λ. Solution is obtained by Newton’s method. If k = 0, the calculation
of λ
1
is skipped. All values are then reported as rates, which are the above numbers divided by the
total exposure.
16 ci — Confidence intervals for means, proportions, and variances
Binomial proportion
Given k successes of n trials, the estimated probability of a success is bp = k/n with standard
error
p
bp(1 − bp)/n. ci calculates the exact (Clopper–Pearson) confidence interval [ p
1
, p
2
] such that
Pr(K ≥ k|p = p
1
) = α/2
and
Pr(K ≤ k|p = p
2
) = α/2
where K is distributed as binomial(n, p). The endpoints may be obtained directly by using Stata’s
invbinomial() function. If k = 0 or k = n, the calculation of the appropriate tail is skipped.
The Wald interval is bp ± z
α/2
p
bp(1 − bp)/n, where z
α/2
is the 1 −α/2 quantile of the standard
normal. The interval is obtained by inverting the acceptance region of the large-sample Wald test of
H
0
: p = p
0
versus the two-sided alternative. That is, the confidence interval is the set of all p
0
such
that
bp − p
0
p
n
−1
bp(1 − bp)
≤ z
α/2
The Wilson interval is a variation on the Wald interval, using the null standard error
p
n
−1
p
0
(1 −p
0
)
in place of the estimated standard error
p
n
−1
bp(1 − bp) in the above expression. Inverting this
acceptance region is more complicated yet results in the closed form
k + z
2
α/2
/2
n + z
2
α/2
±
z
α/2
n
1/2
n + z
2
α/2
(
bp(1 − bp) +
z
2
α/2
4n
)
1/2
The Agresti–Coull interval is basically a Wald interval that borrows its center from the Wilson
interval. Defining
e
k = k + z
2
α/2
/2, en = n + z
2
α/2
, and (hence) ep =
e
k/en, the Agresti–Coull interval
is
ep ± z
α/2
p
ep(1 − ep)/en
When α = 0.05, z
α/2
is near enough to 2 that ep can be thought of as a typical estimate of proportion
where two successes and two failures have been added to the sample (Agresti and Coull 1998).
This typical estimate of proportion makes the Agresti–Coull interval an easy-to-present alternative
for introductory statistics students.
The Jeffreys interval is a Bayesian credible interval and is based on the Jeffreys prior, which
is the Beta(1/2, 1/2) distribution. Assigning this prior to p results in a posterior distribution for
p that is Beta with parameters k + 1/2 and n − k + 1/2. The Jeffreys interval is then taken to
be the 1 − α central posterior probability interval, namely, the α/2 and 1 − α/2 quantiles of the
Beta(k + 1/2, n − k + 1/2) distribution. These quantiles may be obtained directly by using Stata’s
invibeta() function. See [BAYES] bayesstats summary for more details about credible intervals.
Variance and standard deviation
Let X
1
, . . . , X
n
be a random sample and assume that X
i
∼ N(µ, σ
2
). Because (n −1)s
2
/σ
2
∼
χ
2
n−1
, we have Pr{χ
2
n−1,α/2
≤ (n − 1)s
2
/σ
2
≤ χ
2
n−1,1−α/2
} = 1 − α, where χ
2
n−1,α/2
and
χ
2
n−1,1−α/2
are the α/2 and 1 − α/2 quantiles of the χ
2
n−1
distribution. Thus, the normal-based
confidence interval for the population variance σ
2
with 100(1 −α)% confidence level is given by
I
normal
=
"
(n − 1)s
2
χ
2
n−1,1−α/2
,
(n − 1)s
2
χ
2
n−1,α/2
#
ci — Confidence intervals for means, proportions, and variances 17
χ
2
n−1,1−α/2
and χ
2
n−1,α/2
are obtained from Stata as invchi2tail(n-1,0.5*α) and invchi2(n-
1,0.5*α), respectively.
The normal-based confidence interval is very sensitive to minor departures from the normality
assumption, and its performance does not improve with increasing sample size. For scenarios in
which the population distribution is not normal, the actual coverage probability of the normal-based
confidence interval can be drastically lower than the nominal confidence level α.
Bonett (2006) proposed an alternative to the normal-based confidence interval that is nearly exact
under normality and has coverage probability close to 1 −α under moderate nonnormality. It also has
1 −α asymptotic coverage probability for nonnormal distributions with finite fourth moment. Instead
of assuming that X
i
∼ N(µ, σ
2
), Bonett’s approach requires continuous i.i.d. random variables with
finite fourth moment. The variance of s
2
may be expressed as σ
4
{γ
4
− (n − 3)/(n − 1)}/n (see
Casella and Berger [2002, ex. 5.8, 257]), where γ
4
= µ
4
/σ
4
is the kurtosis and µ
4
= E (X
i
− µ)
4
is the population fourth central moment. The variance-stabilizing transformation ln
s
2
and the delta
method can be used to construct an asymptotic 100(1 − α)% confidence interval for σ
2
,
exp
ln
s
2
− z
α/2
se
, exp
ln
s
2
+ z
α/2
se
where se = {bγ
4
− (n − 3)/(n − 1)}/n ≈ Var
ln
s
2
and bγ
4
is an estimate of the kurtosis.
Bonett introduced three adjustments to improve the small-sample properties of the above confidence
interval. First, he swapped the inner and outer denominator in the expression for se and changed it
to {bγ
4
− (n − 3)/n}/(n − 1). This was suggested by Shoemaker (2003) who used it to improve
the small-sample performance of his variance test. Second, with regard to the estimation of kurtosis,
Bonett proposed bγ
4
= n
P
(X
i
− m)
4
/
n
P
X
i
− X
2
o
2
, where m is a trimmed mean with a trim-
proportion equal to 1/
2(n − 4)
1/2
. This kurtosis estimator reduces the negative bias in symmetric
and skewed heavy-tailed distributions. Last, he empirically derived a small-sample correction factor
c = n/(n − z
α/2
) that helps equalize the tail probabilities. These modifications yield
I
Bonett
=
exp
ln
cs
2
− z
α/2
se
, exp
ln
cs
2
+ z
α/2
se
where z
α/2
is the 1 −α/2 quantile of the standard normal and se = c [{bγ
4
− (n − 3)/n}/(n − 1)].
Taking the square root of the endpoints of both intervals gives confidence intervals for the standard
deviation σ.
Edwin Bidwell (E. B.) Wilson (1879–1964) majored in mathematics at Harvard and studied and
taught at Yale and MIT before returning to Harvard in 1922. He worked in mathematics, physics,
and statistics. His method for binomial intervals can be considered a precursor, for a particular
problem, of Neyman’s concept of confidence intervals.
Jerzy Neyman (1894–1981) was born in Bendery, Russia, now Moldavia. He studied and then
taught at Kharkov University, moving from physics to mathematics. In 1921, Neyman moved
to Poland, where he worked in statistics at Bydgoszcz and then Warsaw. Neyman received
a Rockefeller Fellowship to work with Karl Pearson at University College London. There he
collaborated with Egon Pearson, Karl’s son, on the theory of hypothesis testing. Life in Poland
became progressively more difficult, and Neyman returned to UCL to work there from 1934 to 1938.
At this time, he published on the theory of confidence intervals. He then was offered a post in
California at Berkeley, where he settled. Neyman established an outstanding statistics department
and remained highly active in research, including applications in astronomy, meteorology, and
medicine. He was one of the great statisticians of the 20th century.
18 ci — Confidence intervals for means, proportions, and variances
Acknowledgment
We thank Nicholas J. Cox of the Department of Geography at Durham University, UK, who is
coeditor of the Stata Journal and author of Speaking Stata Graphics for his assistance with the
jeffreys and wilson options.
References
Agresti, A., and B. A. Coull. 1998. Approximate is better than “exact” for interval estimation of binomial proportions.
American Statistician 52: 119–126. https://doi.org/10.1080/00031305.1998.10480550.
Bonett, D. G. 2006. Approximate confidence interval for standard deviation of nonnormal distributions. Computational
Statistics and Data Analysis 50: 775–782. https://doi.org/10.1016/j.csda.2004.10.003.
Brown, L. D., T. T. Cai, and A. DasGupta. 2001. Interval estimation for a binomial proportion. Statistical Science
16: 101–133. https://doi.org/10.1214/ss/1009213286.
Campbell, M. J., D. Machin, and S. J. Walters. 2007. Medical Statistics: A Textbook for the Health Sciences. 4th
ed. Chichester, UK: Wiley.
Casella, G., and R. L. Berger. 2002. Statistical Inference. 2nd ed. Pacific Grove, CA: Duxbury.
Clopper, C. J., and E. S. Pearson. 1934. The use of confidence or fiducial limits illustrated in the case of the binomial.
Biometrika 26: 404–413. https://doi.org/10.1093/biomet/26.4.404.
Cook, A. 1990. Sir Harold Jeffreys, 2 April 1891–18 March 1989. Biographical Memoirs of Fellows of the Royal
Society 36: 303–333. https://doi.org/10.1098/rsbm.1990.0034.
Earnest, A. 2017. Essentials of a Successful Biostatistical Collaboration. Boca Raton, FL: CRC Press.
Jeffreys, H. 1946. An invariant form for the prior probability in estimation problems. Proceedings of the Royal Society
of London, Series A 186: 453–461. https://doi.org/10.1098/rspa.1946.0056.
Lash, T. L., T. J. VanderWeele, S. Haneuse, and K. J. Rothman. 2021. Modern Epidemiology. 4th ed. Philadelphia:
Wolters Kluwer.
Lindley, D. V. 2001. Harold Jeffreys. In Statisticians of the Centuries, ed. C. C. Heyde and E. Seneta, 402–405. New
York: Springer.
Miller, R. G., Jr. 1997. Beyond ANOVA: Basics of Applied Statistics. London: Chapman and Hall.
Reid, C. 1982. Neyman—from Life. New York: Springer.
Scheff
´
e, H. 1959. The Analysis of Variance. New York: Wiley.
Shoemaker, L. H. 2003. Fixing the F test for equal variances. American Statistician 57: 105–114.
https://doi.org/10.1198/0003130031441.
Stigler, S. M. 1997. Wilson, Edwin Bidwell. In Leading Personalities in Statistical Sciences: From the Seventeenth
Century to the Present, ed. N. L. Johnson and S. Kotz, 344–346. New York: Wiley.
Ward, B. W. 2019. kg nchs: A command for Korn–Graubard confidence intervals and National Center for Health
Statistics’ Data Presentation Standards for Proportions. Stata Journal 19: 510–522.
Wilson, E. B. 1927. Probable inference, the law of succession, and statistical inference. Journal of the American
Statistical Association 22: 209–212. https://doi.org/10.2307/2276774.
ci — Confidence intervals for means, proportions, and variances 19
Also see
[R] ameans — Arithmetic, geometric, and harmonic means
[R] bitest — Binomial probability test
[R] centile — Report centile and confidence interval
[R] prtest — Tests of proportions
[R] sdtest — Variance-comparison tests
[R] summarize — Summary statistics
[R] ttest — t tests (mean-comparison tests)
[D] pctile — Create variable containing percentiles
Stata, Stata Press, and Mata are registered trademarks of StataCorp LLC. Stata and
Stata Press are registered trademarks with the World Intellectual Property Organization
of the United Nations. StataNow and NetCourseNow are trademarks of StataCorp
LLC. Other brand and product names are registered trademarks or trademarks of their
respective companies. Copyright
c
1985–2023 StataCorp LLC, College Station, TX,
USA. All rights reserved.
®
For suggested citations, see the FAQ on citing Stata documentation.
