Title stata.com
power oneproportion — Power analysis for a one-sample proportion test
Description Quick start Menu Syntax
Options Remarks and examples Stored results Methods and formulas
References Also see
Description
power oneproportion computes sample size, power, or target proportion for a one-sample
proportion test. By default, it computes sample size for given power and the values of the proportion
parameters under the null and alternative hypotheses. Alternatively, it can compute power for given
sample size and values of the null and alternative proportions or the target proportion for given sample
size, power, and the null proportion. For power and sample-size analysis in a cluster randomized
design, see [PSS-2] power oneproportion, cluster. Also see [PSS-2] power for a general introduction
to the power command using hypothesis tests.
Quick start
Sample size for a test of H
0
: π = 0.2 versus H
a
: π 6= 0.2 with null proportion p
0
= 0.2, alternative
proportion p
a
= 0.1, default power of 0.8, and significance level α = 0.05
power oneproportion .2 .1
Same as above, but for p
a
equal to 0.05, 0.075, 0.1, 0.125, and 0.15
power oneproportion .2 (.05(.025).15)
Same as above, but display results as a graph of sample size versus alternative proportion
power oneproportion .2 (.05(.025).15), graph
Sample size for one-sided test with power of 0.9
power oneproportion .2 .1, power(.9) onesided
Sample size for a Wald test instead of the default score test
power oneproportion .2 .1, test(wald)
Power for a sample size of 120
power oneproportion .2 .1, n(120)
Same as above, but for sample sizes of 110, 120, 130, and 140
power oneproportion .2 .1, n(110(10)140)
Effect size and target proportion for p
0
= 0.3, sample size of 75, and power of 0.8
power oneproportion .3, n(75) power(.8)
Menu
Statistics > Power, precision, and sample size
1
2 power oneproportion — Power analysis for a one-sample proportion test
Syntax
Compute sample size
power oneproportion p
0
p
a
, power(numlist) options
Compute power
power oneproportion p
0
p
a
, n(numlist)
options
Compute effect size and target proportion
power oneproportion p
0
, n(numlist) power(numlist)
options
where p
0
is the null (hypothesized) proportion or the value of the proportion under the null hypothesis
and p
a
is the alternative (target) proportion or the value of the proportion under the alternative
hypothesis. p
0
and p
a
may each be specified either as one number or as a list of values in
parentheses (see [U] 11.1.8 numlist).
power oneproportion — Power analysis for a one-sample proportion test 3
options Description
test(test) specify the type of test; default is test(score)
Main
∗
alpha(numlist) significance level; default is alpha(0.05)
∗
power(numlist) power; default is power(0.8)
∗
beta(numlist) probability of type II error; default is beta(0.2)
∗
n(numlist) sample size; required to compute power or effect size
nfractional allow fractional sample size
∗
diff(numlist) difference between the alternative proportion and the null
proportion, p
a
− p
0
; specify instead of the
alternative proportion p
a
critvalues show critical values for the binomial test
continuity apply continuity correction to the normal approximation
of the discrete distribution
direction(upper|lower) direction of the effect for effect-size determination; default is
direction(upper), which means that the postulated value
of the parameter is larger than the hypothesized value
onesided one-sided test; default is two sided
parallel treat number lists in starred options or in command arguments as
parallel when multiple values per option or argument are
specified (do not enumerate all possible combinations of values)
Table
no
table
(tablespec)
suppress table or display results as a table;
see [PSS-2] power, table
saving(filename
, replace
) save the table data to filename; use replace to overwrite
existing filename
Graph
graph
(graphopts)
graph results; see [PSS-2] power, graph
Iteration
init(#) initial value for sample size or proportion
iterate(#) maximum number of iterations; default is iterate(500)
tolerance(#) parameter tolerance; default is tolerance(1e-12)
ftolerance(#) function tolerance; default is ftolerance(1e-12)
no
log suppress or display iteration log
no
dots suppress or display iterations as dots
cluster perform computations for a CRD;
see [PSS-2] power oneproportion, cluster
notitle suppress the title
∗
Specifying a list of values in at least two starred options, or at least two command arguments, or at least one
starred option and one argument results in computations for all possible combinations of the values; see
[U] 11.1.8 numlist. Also see the parallel option.
collect is allowed; see [U] 11.1.10 Prefix commands.
cluster and notitle do not appear in the dialog box.
4 power oneproportion — Power analysis for a one-sample proportion test
test Description
score score test; the default
wald Wald test
binomial binomial test
test() does not appear in the dialog box. The dialog box selected is determined by the test() specification.
where tablespec is
column
:label
column
:label
. . .
, tableopts
column is one of the columns defined below, and label is a column label (may contain quotes and
compound quotes).
column Description Symbol
alpha significance level α
alpha a observed significance level α
a
power power 1 −β
beta type II error probability β
N number of subjects N
delta effect size δ
p0 null proportion p
0
pa alternative proportion p
a
diff difference between the alternative and null p
a
− p
0
proportions
C l lower critical value C
l
C u upper critical value C
u
target target parameter; synonym for pa
all display all supported columns
Column beta is shown in the default table in place of column power if specified.
Column diff is shown in the default table if specified.
Columns alpha a, C l, and C u are available when the test(binomial) option is specified.
Columns C l and C u are shown in the default table, if the critvalues option is specified.
Options
test(test) specifies the type of the test for power and sample-size computations. test is one of
score, wald, or binomial.
score requests computations for the score test. This is the default test.
wald requests computations for the Wald test. This corresponds to computations using the value
of the alternative proportion instead of the default null proportion in the formula for the standard
error of the estimator of the proportion.
binomial requests computations for the binomial test. The computation using the binomial
distribution is not available for sample-size and effect-size determinations; see example 7 for
details. Iteration options are not allowed with this test.
power oneproportion — Power analysis for a one-sample proportion test 5
Main
alpha(), power(), beta(), n(), nfractional; see [PSS-2] power. The nfractional option is
allowed only for sample-size determination.
diff(numlist) specifies the difference between the alternative proportion and the null proportion,
p
a
−p
0
. You can specify either the alternative proportion p
a
as a command argument or the difference
between the two proportions in diff(). If you specify diff(#), the alternative proportion is
computed as p
a
= p
0
+ #. This option is not allowed with the effect-size determination.
critvalues requests that the critical values be reported when the computation is based on the
binomial distribution.
continuity requests that continuity correction be applied to the normal approximation of the discrete
distribution. continuity cannot be specified with test(binomial).
direction(), onesided, parallel; see [PSS-2] power.
Table
table, table(), notable; see [PSS-2] power, table.
saving(); see [PSS-2] power.
Graph
graph, graph(); see [PSS-2] power, graph. Also see the column table for a list of symbols used by
the graphs.
Iteration
init(#) specifies the initial value of the sample size for the sample-size determination or the initial
value of the proportion for the effect-size determination.
iterate(), tolerance(), ftolerance(), log, nolog, dots, nodots; see [PSS-2] power.
The following options are available with power oneproportion but are not shown in the dialog
box:
cluster; see [PSS-2] power oneproportion, cluster.
notitle; see [PSS-2] power.
Remarks and examples stata.com
Remarks are presented under the following headings:
Introduction
Using power oneproportion
Computing sample size
Computing power
Computing effect size and target proportion
Performing hypothesis tests on proportion
Video examples
This entry describes the power oneproportion command and the methodology for power and
sample-size analysis for a one-sample proportion test. See [PSS-2] Intro (power) for a general
introduction to power and sample-size analysis and [PSS-2] power for a general introduction to the
power command using hypothesis tests. Also see [PSS-2] power oneproportion, cluster for power
and sample-size analysis in a cluster randomized design.
6 power oneproportion — Power analysis for a one-sample proportion test
Introduction
There are many examples of studies where a researcher would like to compare an observed
proportion with a hypothesized proportion. A political campaign might like to know if the proportion
of a country’s population that supports a new legislative initiative is greater than 50%. A veterinary
drug manufacturer might test a new topical treatment to kill fleas on dogs. It would like to know the
sample size necessary to demonstrate that the treatment is effective in ridding at least 80% of the test
dogs of fleas. The Nevada Gaming Control Board might test a Las Vegas casino’s slot machines to
verify that it meets the statutory minimum payout percentage of 75%. The board would like to know
the number of “pulls” necessary to reject the one-sided null hypothesis that the payout percentage is
less than 75%.
The analysis of proportions is carried out in experiments or observational studies where the response
variable is binary. Each observation is an outcome from a Bernoulli trial with a fixed probability p
of observing an event of interest in a population. Hypothesis testing of binomial outcomes relies on
a set of assumptions: 1) Bernoulli outcome is observed a fixed number of times; 2) the probability p
is fixed across all trials; and 3) individual trials are independent.
This entry describes power and sample-size analysis for the inference about the population proportion
performed using hypothesis testing. Specifically, we consider the null hypothesis H
0
: p = p
0
versus
the two-sided alternative hypothesis H
a
: p 6= p
0
, the upper one-sided alternative H
a
: p > p
0
, or the
lower one-sided alternative H
a
: p < p
0
.
Two common hypothesis tests for a one-sample proportion are the small-sample binomial test and
the asymptotic (large-sample) normal test. The binomial test is based on the binomial distribution,
the exact sampling distribution, of the test statistic and is commonly known as an “exact binomial”
test. The asymptotic normal test is based on the large-sample normal approximation of the sampling
distribution of the test statistic and is often referred to as a z test.
power oneproportion provides power and sample-size analysis for both the binomial and a
large-sample z test of a one-sample proportion.
Using power oneproportion
power oneproportion computes sample size, power, or target proportion for a one-sample
proportion test. All computations are performed for a two-sided hypothesis test where, by default, the
significance level is set to 0.05. You may change the significance level by specifying the alpha()
option. You can specify the onesided option to request a one-sided test.
power oneproportion performs power analysis for three different tests, which can be specified
within the test() option. The default is a large-sample score test (test(score)), which approximates
the sampling distribution of the test statistic by the standard normal distribution. You may instead
request computations based on a large-sample Wald test by specifying the test(wald) option.
For power determination, you can also request the small-sample binomial test by specifying the
test(binomial) option. The binomial test is not available for the sample-size and effect-size
determinations; see example 7 for details.
To compute sample size, you must specify the proportions under the null and alternative hypotheses,
p
0
and p
a
, respectively, and, optionally, the power of the test in the power() option. The default
power is set to 0.8.
To compute power, you must specify the sample size in the n() option and the proportions under
the null and alternative hypotheses, p
0
and p
a
, respectively.
power oneproportion — Power analysis for a one-sample proportion test 7
Instead of the alternative proportion p
a
, you may specify the difference p
a
− p
0
between the
alternative proportion and the null proportion in the diff() option when computing sample size or
power.
To compute effect size, the difference between the alternative and null proportions, and target
proportion, you must specify the sample size in the n() option, the power in the power() option,
the null proportion p
0
, and, optionally, the direction of the effect. The direction is upper by default,
direction(upper), which means that the target proportion is assumed to be larger than the specified
null value. You can change the direction to lower, which means that the target proportion is assumed
to be smaller than the specified null value, by specifying the direction(lower) option.
By default, the computed sample size is rounded up. You can specify the nfractional option
to see the corresponding fractional sample size; see Fractional sample sizes in [PSS-4] Unbalanced
designs for an example. The nfractional option is allowed only for sample-size determination.
Some of power oneproportion’s computations require iteration. For example, for a large-sample
z test, sample size for a two-sided test is obtained by iteratively solving a nonlinear power equation.
The default initial value for the sample size for the iteration procedure is obtained using a closed-form
one-sided formula. If desired, it may be changed by specifying the init() option. See [PSS-2] power
for the descriptions of other options that control the iteration procedure.
In the following sections, we describe the use of power oneproportion accompanied with
examples for computing sample size, power, and target proportion.
Computing sample size
To compute sample size, you must specify the proportions under the null and alternative hypotheses,
p
0
and p
a
, respectively, and, optionally, the power of the test in the power() option. A default power
of 0.8 is assumed if power() is not specified.
Example 1: Sample size for a one-sample proportion test
Consider a study of osteoporosis in postmenopausal women from Chow et al. (2018, 45). The
term “osteoporosis” refers to the decrease in bone mass that is most prevalent in postmenopausal
women. Females diagnosed with osteoporosis have vertebral bone density more than 10% below the
average bone density of women with similar demographic characteristics such as age, height, weight,
and race.
The World Health Organization (WHO) defines osteoporosis as having the bone density value that
is smaller than 2.5 standard deviations below the peak bone mass levels in young women. Suppose
investigators wish to assess the effect of a new treatment on increasing the bone density for women
diagnosed with osteoporosis. The treatment is deemed successful if a subject’s bone density improves
by more than one standard deviation of her measured bone density.
Suppose that previous studies have reported a response rate of 30% for women with increased
bone density after treatment. Investigators expect the new treatment to generate a higher response
rate of roughly 50%. The goal is to obtain the minimum required sample size to detect an alternative
proportion of 0.5 using the test of H
0
: p = 0.3 versus H
a
: p 6= 0.3 with 80% power and 5%
significance level. To compute sample size, we specify the null and alternative proportions after the
command name:
8 power oneproportion — Power analysis for a one-sample proportion test
. power oneproportion 0.3 0.5
Performing iteration ...
Estimated sample size for a one-sample proportion test
Score z test
H0: p = p0 versus Ha: p != p0
Study parameters:
alpha = 0.0500
power = 0.8000
delta = 0.2000
p0 = 0.3000
pa = 0.5000
Estimated sample size:
N = 44
We find that at least 44 subjects are needed to detect a change in proportion from 0.3 to 0.5 with
80% power using a 5%-level two-sided test.
Example 2: Specifying the difference between proportions
Instead of the alternative proportion, we can specify the difference of 0.05 −0.03 = 0.2 between
the alternative proportion and the null proportion in the diff() option and obtain the same results:
. power oneproportion 0.3, diff(0.2)
Performing iteration ...
Estimated sample size for a one-sample proportion test
Score z test
H0: p = p0 versus Ha: p != p0
Study parameters:
alpha = 0.0500
power = 0.8000
delta = 0.2000
p0 = 0.3000
pa = 0.5000
diff = 0.2000
Estimated sample size:
N = 44
The difference between proportions is now also displayed in the output.
Example 3: Wald test
The default computation is based on a score test and thus uses the null proportion as the estimate
of the true proportion in the formula for the standard error. We can request the computation based
on a Wald test by specifying the test(wald) option. In this case, the alternative proportion will be
used as an estimate of the true proportion in the formula for the standard error.
power oneproportion — Power analysis for a one-sample proportion test 9
. power oneproportion 0.3 0.5, test(wald)
Performing iteration ...
Estimated sample size for a one-sample proportion test
Wald z test
H0: p = p0 versus Ha: p != p0
Study parameters:
alpha = 0.0500
power = 0.8000
delta = 0.2000
p0 = 0.3000
pa = 0.5000
Estimated sample size:
N = 50
We find that the required sample size increases to 50 subjects.
Computing power
To compute power, you must specify the sample size in the n() option and the proportions under
the null and alternative hypotheses, p
0
and p
a
, respectively.
Example 4: Power of a one-sample proportion test
Continuing with example 1, we will suppose that we are designing a new study and anticipate to
obtain a sample of 30 subjects. To compute the power corresponding to this sample size given the
study parameters from example 1, we specify the sample size of 30 in the n() option:
. power oneproportion 0.3 0.5, n(30)
Estimated power for a one-sample proportion test
Score z test
H0: p = p0 versus Ha: p != p0
Study parameters:
alpha = 0.0500
N = 30
delta = 0.2000
p0 = 0.3000
pa = 0.5000
Estimated power:
power = 0.6534
As expected, with a smaller sample size, we achieve a lower power (only 65.34%).
10 power oneproportion — Power analysis for a one-sample proportion test
Example 5: Multiple values of study parameters
To see the effect of sample size on power, we can specify a range of sample sizes in the n()
option.
. power oneproportion 0.3 0.5, n(40(1)50)
Estimated power for a one-sample proportion test
Score z test
H0: p = p0 versus Ha: p != p0
alpha power N delta p0 pa
.05 .7684 40 .2 .3 .5
.05 .7778 41 .2 .3 .5
.05 .787 42 .2 .3 .5
.05 .7958 43 .2 .3 .5
.05 .8043 44 .2 .3 .5
.05 .8124 45 .2 .3 .5
.05 .8203 46 .2 .3 .5
.05 .8279 47 .2 .3 .5
.05 .8352 48 .2 .3 .5
.05 .8422 49 .2 .3 .5
.05 .849 50 .2 .3 .5
As expected, power is an increasing function of the sample size.
For multiple values of parameters, the results are automatically displayed in a table, as we see
above. For more examples of tables, see [PSS-2] power, table. If you wish to produce a power plot,
see [PSS-2] power, graph.
Example 6: Sign test
We can use power oneproportion to perform power and sample-size analysis for a nonparametric
sign test comparing the median of a sample with a reference value. The sign test for comparing a
median is simply a test of a binomial proportion with the reference (null) value of 0.5, H
0
: p = 0.5.
For example, consider a study similar to the one described in example 1. Suppose we want to test
whether the median bone density exceeds a threshold value in a population of females who received
a certain treatment. This is equivalent to testing whether the proportion p of bone-density values
exceeding the threshold is greater than 0.5, that is, H
0
: p = 0.5 versus H
a
: p > 0.5. Suppose that
from previous studies such proportion was estimated to be 0.7. We anticipate to enroll 30 subjects
and would like to compute the corresponding power of an upper one-sided small-sample binomial
test to detect the change in proportion from 0.5 to 0.7.
power oneproportion — Power analysis for a one-sample proportion test 11
. power oneproportion 0.5 0.7, n(30) test(binomial) onesided
Estimated power for a one-sample proportion test
Binomial test
H0: p = p0 versus Ha: p > p0
Study parameters:
alpha = 0.0500
N = 30
delta = 0.2000
p0 = 0.5000
pa = 0.7000
Estimated power and alpha:
power = 0.7304
actual alpha = 0.0494
For a sample size of 30 subjects, we obtain a power of 73% to detect the difference of 0.2 between
the alternative and null values. In addition to power, power oneproportion also displays the actual
(observed) significance level, which is 0.0494 in our example and is very close to the specified
significance level of 0.05.
When the sampling distribution of the test statistic is discrete such as for the binomial test, the
specified nominal significance level may not be possible to precisely achieve, because the space
of the observed significance levels is discrete. As such, power oneproportion also displays the
observed significance level given the specified sample size, power, and other study parameters. Also
see example 7.
Example 7: Saw-toothed power function
In example 6, we briefly described one issue arising with power and sample-size analysis for the
binomial test. The observed significance levels are discrete because the binomial sampling distribution
of the test statistic is discrete. Another related issue arising because of the discrete nature of the
sampling distribution is the nonmonotonic relationship between power and sample size—as the sample
size increases, the corresponding power may not necessarily increase. The power function may have
a so-called saw-toothed shape (Chernick and Liu 2002), where it increases initially, then drops, then
increases again, and so on. See figure 1 below for an example.
To demonstrate the issue, we return to example 5 and plot powers for a range of sample size
values between 45 and 60. We specify the graph() option to produce a graph and the table()
option to produce a table; see [PSS-2] power, graph and [PSS-2] power, table for more details about
the graphical and tabular outputs from power. Within graph(), we request that the reference line be
plotted on the y axis at a power of 0.8 and that the data points bee labeled with the corresponding
sample sizes. Within table(), we specify the formats() suboption to display only three digits after
the decimal point for the power and alpha a columns. We also specify the critvalues option to
display columns containing critical values in the table.
12 power oneproportion — Power analysis for a one-sample proportion test
. power oneprop 0.3 0.5, n(45(1)60) test(binomial) critvalues
> table(, formats(alpha_a "%7.3f" power "%7.3f"))
> graph(yline(0.8) plotopts(mlabel(N)))
Estimated power for a one-sample proportion test
Binomial test
H0: p = p0 versus Ha: p != p0
alpha alpha_a power N delta p0 pa C_l C_u
.05 0.034 0.724 45 .2 .3 .5 7 21
.05 0.035 0.769 46 .2 .3 .5 7 21
.05 0.037 0.809 47 .2 .3 .5 7 21
.05 0.026 0.765 48 .2 .3 .5 7 22
.05 0.042 0.804 49 .2 .3 .5 8 22
.05 0.031 0.760 50 .2 .3 .5 8 23
.05 0.031 0.799 51 .2 .3 .5 8 23
.05 0.033 0.834 52 .2 .3 .5 8 23
.05 0.037 0.795 53 .2 .3 .5 9 24
.05 0.037 0.830 54 .2 .3 .5 9 24
.05 0.038 0.860 55 .2 .3 .5 9 24
.05 0.028 0.825 56 .2 .3 .5 9 25
.05 0.043 0.855 57 .2 .3 .5 10 25
.05 0.044 0.881 58 .2 .3 .5 10 25
.05 0.032 0.851 59 .2 .3 .5 10 26
.05 0.033 0.877 60 .2 .3 .5 10 26
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
.7
.75
.8
.85
.9
Power (1-β)
45 50 55 60
Sample size (N)
Parameters: α = .05, δ = .2, p
0
= .3, p
a
= .5
Binomial test
H
0
: p = p
0
versus H
a
: p ≠ p
0
Estimated power for a one-sample proportion test
Figure 1. Saw-toothed power function
The power is not a monotonic function of the sample size. Also from the table, we can see that all
the observed significance levels are smaller than the specified level of 0.05.
To better understand what is going on, we will walk through the steps of power determination.
First, the critical values are determined as the minimum value C
l
and the maximum value C
u
between
0 and n that satisfy the following inequalities,
Pr(X ≤ C
l
|p = p
0
) ≤ α/2 and Pr(X ≥ C
u
|p = p
0
) ≤ α/2
power oneproportion — Power analysis for a one-sample proportion test 13
where the number of successes X has a binomial distribution with the total number of trials n and a
probability of a success in a single trial p, X ∼ Bin(n, p). The power is then computed as the sum
of the above two probabilities with p = p
a
.
For example, let’s compute the power for the first row of the table. The sample size is 45, the lower
critical value is 7, and the upper critical value is 21. We use the probability functions binomial()
and binomialtail() to compute the respective lower- and upper-tailed probabilities of the binomial
distribution.
. di "Lower tail: " binomial(45,7,0.3)
Lower tail: .0208653
. di "Upper tail: " binomialtail(45,21,0.3)
Upper tail: .01352273
. di "Obs. level: " binomial(45,7,0.3) + binomialtail(45,21,0.3)
Obs. level: .03438804
. di "Power: " binomial(45,7,0.5) + binomialtail(45,21,0.5)
Power: .7242594
Each of the tails is less than 0.025 (α/2 = 0.05/2 = 0.025). The observed significance level and
power match the results from the first row of the table.
Now let’s increase the lower critical value by one, C
l
= 8, and decrease the upper critical value
by one, C
u
= 20:
. di "Lower tail: " binomial(45,8,0.3)
Lower tail: .04711667
. di "Upper tail: " binomialtail(45,20,0.3)
Upper tail: .02834511
Each of the tail probabilities now exceeds 0.025. If we could use values between 7 and 8 and between
20 and 21, we could match the tails exactly to 0.025, and then the monotonicity of the power function
would be preserved. This is impossible for the binomial distribution (or any discrete distribution)
because the number of successes must be integers.
Because of the saw-toothed nature of the power curve, obtaining an optimal sample size becomes
tricky. If we wish to have power of 80%, then from the above table and graph, we see that potential
sample sizes are 47, 49, 52, 54, and so on. One may be tempted to choose the smallest sample
size for which the power is at least 80%. This, however, would not guarantee that the power is at
least 80% for any larger sample size. Instead, Chernick and Liu (2002) suggest selecting the smallest
sample size after which the troughs of the power curve do not go below the desired power. Following
this recommendation in our example, we would pick a sample size of 54, which corresponds to the
observed significance level of 0.037 and power of 0.83.
In the above, we showed the power curve for the sample sizes between 45 and 60. It may be a
good idea to also look at the power plot for larger sample sizes to verify that the power continues to
increase and does not drop below the desired power level.
Computing effect size and target proportion
In an analysis of a one-sample proportion, the effect size δ is often defined as the difference
between the alternative proportion and the null proportion, δ = p
a
− p
0
.
Sometimes, we may be interested in determining the smallest effect and the corresponding alternative
or target proportion that yield a statistically significant result for prespecified sample size and power.
In this case, power, sample size, and null proportion must be specified. In addition, you must also
14 power oneproportion — Power analysis for a one-sample proportion test
decide on the direction of the effect: upper, meaning p
a
> p
0
, or lower, meaning p
a
< p
0
. The
direction may be specified in the direction() option; direction(upper) is the default.
Example 8: Minimum detectable value of the proportion
Continuing with example 4, we may also be interested to find the minimum value of the proportion
that can be detected with a power of 80% given a sample of 30 subjects. To compute this, after the
command name, we specify the null proportion of 0.3, sample size n(30), and power power(0.8):
. power oneproportion 0.3, n(30) power(0.8)
Performing iteration ...
Estimated target proportion for a one-sample proportion test
Score z test
H0: p = p0 versus Ha: p != p0; pa > p0
Study parameters:
alpha = 0.0500
power = 0.8000
N = 30
p0 = 0.3000
Estimated effect size and target proportion:
delta = 0.2406
pa = 0.5406
The smallest detectable value of the proportion is 0.54.
In the above, we assumed the effect to be in the upper direction, p
a
> p
0
. We can obtain the
results in the lower direction by specifying the direction(lower) option.
Performing hypothesis tests on proportion
In this section, we briefly demonstrate how you can test hypotheses about proportions; see [R] prtest
and [R] bitest for details. Suppose we wish to test the hypothesis that the proportion is different from
a reference value on the collected data. We can use the prtest command or the bitest command
to do this.
Example 9: Testing for proportion
We use lbw.dta, which contains data on birthweights of infants from a sample of 189 females.
One of the variables in the dataset is variable ui, which records the presence or absence of uterine
irritability. Although the real objective of this study is different, suppose we wish to test the null
hypothesis that the proportion of women in a sample who experience uterine irritability is equal to
0.20. We can use the prtest command to perform a large-sample test of a single proportion.
power oneproportion — Power analysis for a one-sample proportion test 15
. use https://www.stata-press.com/data/r18/lbw
(Hosmer & Lemeshow data)
. prtest ui==0.2
One-sample test of proportion Number of obs = 189
Variable Mean Std. err. [95% conf. interval]
ui .1481481 .0258404 .0975019 .1987944
p = proportion(ui) z = -1.7821
H0: p = 0.2
Ha: p < 0.2 Ha: p != 0.2 Ha: p > 0.2
Pr(Z < z) = 0.0374 Pr(|Z| > |z|) = 0.0747 Pr(Z > z) = 0.9626
We do not have statistical evidence to reject the null hypothesis of H
0
: p = 0.2 versus a two-sided
alternative H
a
: p 6= 0.2 at least at the 5% significance level; the p-value = 0.0747 > 0.05.
If our true objective were to study uterine irritability in the population of females, we would have
performed the corresponding power and sample-size analysis before collecting the data. For example,
using the estimates of lbw.dta, we can use power oneproportion to compute the required sample
size for a 5%-level two-sided large-sample z test to detect the change in proportion from the reference
value of 0.2 to approximately 0.148 with a power of, say, 80%:
. power oneproportion 0.2 0.148
Performing iteration ...
Estimated sample size for a one-sample proportion test
Score z test
H0: p = p0 versus Ha: p != p0
Study parameters:
alpha = 0.0500
power = 0.8000
delta = -0.0520
p0 = 0.2000
pa = 0.1480
Estimated sample size:
N = 434
We find that we need 434 subjects, many more than the current sample of 189, to detect the specified
change in proportions.
Video examples
Sample-size calculation for comparing a sample proportion to a reference value
Power calculation for comparing a sample proportion to a reference value
Minimum detectable effect size for comparing a sample proportion to a reference value using Stata
16 power oneproportion — Power analysis for a one-sample proportion test
Stored results
power oneproportion stores the following in r():
Scalars
r(alpha) significance level
r(alpha a) actual significance level of the binomial method
r(power) power
r(beta) probability of a type II error
r(delta) effect size
r(N) sample size
r(nfractional) 1 if nfractional is specified, 0 otherwise
r(onesided) 1 for a one-sided test, 0 otherwise
r(p0) proportion under the null hypothesis
r(pa) proportion under the alternative hypothesis
r(diff) difference between the alternative and null proportions
r(C l) lower critical value of the binomial distribution
r(C u) upper critical value of the binomial distribution
r(continuity) 1 if continuity correction is used, 0 otherwise
r(separator) number of lines between separator lines in the table
r(divider) 1 if divider is requested in the table, 0 otherwise
r(init) initial value for sample size or proportion
r(maxiter) maximum number of iterations
r(iter) number of iterations performed
r(tolerance) requested parameter tolerance
r(deltax) final parameter tolerance achieved
r(ftolerance) requested distance of the objective function from zero
r(function) final distance of the objective function from zero
r(converged) 1 if iteration algorithm converged, 0 otherwise
Macros
r(type) test
r(method) oneproportion
r(test) score, wald, or binomial
r(direction) upper or lower
r(columns) displayed table columns
r(labels) table column labels
r(widths) table column widths
r(formats) table column formats
Matrices
r(pss table) table of results
Methods and formulas
Let x
1
, . . . , x
n
be a sequence of n independent and identically distributed Bernoulli random
variates. Let x
i
= 1 denote a success and x
i
= 0 denote a failure. Let P (x
i
= 1) = p denote the
probability of a success in the population. Each individual observation is a Bernoulli trial with a
success probability p, which implies that the sum X =
P
n
i=1
x
i
has a binomial distribution with
mean np and standard deviation
p
np(1 −p). Let
bp =
1
n
n
X
i=1
x
i
and se(bp) =
r
bp(1 − bp)
n
denote the sample proportion and its standard error, respectively. Let p
0
and p
a
denote the null and
alternative values of the proportion parameter, respectively.
A one-sample proportion test involves testing the null hypothesis H
0
: p = p
0
versus the two-sided
alternative hypothesis H
a
: p 6= p
0
, the upper one-sided alternative H
a
: p > p
0
, or the lower one-sided
alternative H
a
: p < p
0
.
power oneproportion — Power analysis for a one-sample proportion test 17
If the nfractional option is not specified, the computed sample size is rounded up.
The following formulas are based on Chow et al. (2018).
Methods and formulas are presented under the following headings:
Large-sample normal approximation
Binomial test
Large-sample normal approximation
For a large sample, the distribution of the sample proportion bp may be approximated by the normal
distribution with mean p and variance p(1 − p)/n. Two test statistics are considered: the score test
statistic z = (bp −p
0
)/
p
p
0
(1 −p
0
)/n and the Wald test statistic z = (bp −p
0
)/
p
bp(1 − bp)/n. The
score test statistic uses the null value of the proportion to construct the standard error, which leads
to its sampling distribution being closer to the standard normal distribution than if the Wald statistic
were used (Agresti [2013, 13]).
Let α be the significance level, β be the probability of a type II error, and z
1−α
and z
β
be the (1−α)th
and the βth quantiles of the standard normal distribution. Also let η =
p
{p
0
(1 −p
0
)}/{p
a
(1 −p
a
)}.
The power π = 1 − β of the score z test is computed using
π =
Φ
√
n(p
a
−p
0
)−c
√
p
a
(1−p
a
)
− z
1−α
η
upper one sided
Φ
−
√
n(p
a
−p
0
)−c
√
p
a
(1−p
a
)
− z
1−α
η
lower one sided
Φ
√
n(p
a
−p
0
)−c
√
p
a
(1−p
a
)
− z
1−α/2
η
+ Φ
−
√
n(p
a
−p
0
)−c
√
p
a
(1−p
a
)
− z
1−α/2
η
two sided
(1)
where Φ(·) is the cdf of the standard normal distribution, and c is the normal-approximation continuity
correction: c = 1/(2
√
n) if the continuity option is specified, and c = 0 otherwise.
The power of the Wald z test can be obtained from (1) by replacing the term p
0
(1 − p
0
) in η
with p
a
(1 −p
a
) so that η = 1.
The sample size n for a one-sided test is computed using
n =
(
z
1−α
p
p
0
(1 − p
0
) + z
1−β
p
p
a
(1 − p
a
)
δ
)
2
If the continuity option is specified, the sample size n
c
for a one-sided test is computed as
n
c
=
n
4
1 +
s
1 +
2
n|p
a
− p
0
|
!
2
where n is the sample size computed without the correction (Fleiss, Levin, and Paik 2003; Levin and
Chen 1999).
The sample size for a two-sided test and minimum detectable value of the proportion are computed
iteratively using the corresponding power equation from (1).
18 power oneproportion — Power analysis for a one-sample proportion test
Binomial test
Power of the binomial test is computed using the binomial (exact) sampling distribution of the
test statistic. Consider a one-sided test given by
H
0
: p = p
0
versus H
a
: p > p
0
Let X denote the number of successes in the sample. The null hypothesis is rejected if X is
greater than a critical value k such that the resulting p-value is less than or equal to the significance
level α.
The p-value for testing the above one-sided hypothesis can be obtained from the following equation:
P (X ≥ k; n, p
0
) =
n
X
i=k
n
i
p
i
0
(1 − p
0
)
n−i
The p-value for testing the two-sided hypothesis H
a
: p 6= p
0
is given by
2 × min {P (X ≥ k; n, p
0
), P (X ≤ k; n, p
0
)}
For a one-sided test, the power of the test is computed from the following nonlinear equation:
π = 1 − β =
n
X
k=0
n
k
p
k
a
(1 − p
a
)
n−k
I {P (X ≥ k; n, p
0
) ≤ α}
Power for a two-sided test can be obtained by replacing the indicator function above with
I [2 × min {P (X ≥ k; n, p
0
), P (X ≤ k; n, p
0
)} ≤ α].
The computational details may be found in Krishnamoorthy and Peng (2007).
References
Agresti, A. 2013. Categorical Data Analysis. 3rd ed. Hoboken, NJ: Wiley.
Chernick, M. R., and C. Y. Liu. 2002. The saw-toothed behavior of power versus sample size and
software solutions: Single binomial proportion using exact methods. American Statistician 56: 149–155.
https://doi.org/10.1198/000313002317572835.
Chow, S.-C., J. Shao, H. Wang, and Y. Lokhnygina. 2018. Sample Size Calculations in Clinical Research. 3rd ed.
Boca Raton, FL: CRC Press.
Fleiss, J. L., B. Levin, and M. C. Paik. 2003. Statistical Methods for Rates and Proportions. 3rd ed. New York:
Wiley.
Krishnamoorthy, K., and J. Peng. 2007. Some properties of the exact and score methods for binomial propor-
tion and sample size calculation. Communications in Statistics—Simulation and Computation 36: 1171–1186.
https://doi.org/10.1080/03610910701569218.
Levin, B., and X. Chen. 1999. Is the one-half continuity correction used once or twice to derive a well-known
approximate sample size formula to compare two independent binomial distributions? American Statistician 53:
62–66. https://doi.org/10.1080/00031305.1999.10474431.
power oneproportion — Power analysis for a one-sample proportion test 19
Also see
[PSS-2] power oneproportion, cluster — Power analysis for a one-sample proportion test, CRD
[PSS-2] power — Power and sample-size analysis for hypothesis tests
[PSS-2] power, graph — Graph results from the power command
[PSS-2] power, table — Produce table of results from the power command
[PSS-5] Glossary
[ADAPT] gsdesign oneproportion — Group sequential design for a one-sample proportion test
[R] bitest — Binomial probability test
[R] prtest — Tests of proportions
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.
