Estimation and Confidence Intervals

Fall 2001 Professor Paul Glasserman

B6014: Managerial Statistics 403 Uris Hall

Properties of Point Estimates

1. We have already encountered two point estimators: the sample mean X is an estimator
of the population mean µ, and the sample proportion p̂ is an estimator of the population
proportion p. These are called point estimates in contrast to interval estimates. A
point estimate is a single number (such as X or p̂) whereas an interval estimate gives a
range likely to contain the true value of the unknown parameter.

2. What makes one estimator better than another? Is there some sense in which X and p̂
are the best possible estimates of µ and p, given the available data? We briefly discuss
the relevant considerations in addressing these questions.

3. There are two distinct considerations to keep in mind in evaluating an estimator: bias
and variability. Variability is measured by the standard error of the estimator, and we
have encountered this already a couple of times. We have not explicitly discussed bias
previously, so we focus on it now.

4. In everyday usage, bias has many meanings, but in statistics it has just one: the bias of an
estimator is the difference between its expected value and the quantity being estimated:

Bias(estimator) = E[estimator] − quantity being estimated.

An estimator is called unbiased if this difference is zero; i.e., if

E[estimator] = quantity being estimated.

5. The intuitive content of this is that an unbiased estimator is one that does not systemat-
ically overestimate or underestimate the target quantity.

1
 6. We have already met two unbiased estimators: the sample mean X is an unbiased es-
timator of the population mean µ because E[X] = µ; the sample proportion p̂ is an
unbiased estimator of the population proportion p because E[p̂] = p. Each of these esti-
mators is correct “on average,” in the sense that neither systematically overestimates or
underestimates.
7. We have discussed estimation of a population mean µ but not estimation of a popula-
tion variance σ 2 . We estimate σ 2 from a random sample X1 , . . . , Xn using the sample
variance
1
n
s2 = (Xi − X)2 . (1)
n − 1 i=1
It is a (non-obvious) mathematical fact that E[s2 ] = σ 2 ; i.e., the sample variance is
unbiased. That’s why we divide by n − 1; if we divided by n, we would get a slightly
smaller estimate and indeed one that is biased low. Notice, however, that even if we
divided by n, the bias would vanish as n becomes large because (n − 1)/n approaches 1
as n increases.
8. The sample standard deviation


 1
n
s= (Xi − X)2
n − 1 i=1

is not an unbiased estimator of the population standard deviation σ. It is biased low,

because E[s] < σ (this isn’t obvious, but it’s true). However, the bias vanishes as the
sample size increases, so we use this estimator all the time.
9. The preceeding example shows that a biased estimator is not necessarily a bad estimator,
so long as the bias is small and disappears as the sample size increases.

Confidence Intervals for the Population Mean

1. We have seen that the sample mean X is an unbiased estimator of the population mean
µ. This means that X is accurate, on average; but of course for any paricular data set
X1 , . . . , Xn , the sample mean may be higher or lower than the true value µ. The purpose
of a confidence interval is to supplement the point estimate X with information about
the uncertainty in this estimate.
2. A confidence interval for µ takes the form X̄ ± ∆, with the range ∆ chosen so that there
is, e.g., a 95% probability that the error X̄ − µ is less than ±∆; i.e.,
P (−∆ < X − µ < ∆) = .95.
√
We know that X − µ is approximately normal with mean 0 and standard deviation σ/ n.
So, to capture the middle 95% of the distribution we should set
σ
∆ = 1.96 √ .
n

2
 We can now say that we are 95% confident that the true mean lies in the interval
σ σ
(X − 1.96 √ , X + 1.96 √ ).
n n
This is a 95% confidence interval for µ.

3. In deriving this interval, we assumed that X is approximately normal N (µ, σ 2 /n). This
is valid if the underlying population (from which X1 , . . . , Xn are drawn) is normal or else
if the sample size is sufficiently large (say n ≥ 30).

4. Example: Apple Tree Supermarkets is considering opening a new store at a certain loca-
tion, but wants to know if average weekly sales will reach $250,000. Apple Tree estimates
weekly gross sales at nearby stores by sending field workers to collect observations. The
field workers collect 40 samples and arrive at a sample mean of $263,590. Suppose the
standard deviation of these samples is $42,000. Find a 90% confidence interval for µ.
If we ask for a 90% confidence interval, we should use 1.645 instead of 1.96 as the multi-
plier. This is interval is thus given by
√ √
(263590 − 1.645(42000/ 40), 263590 + 1.645(42000/ 40))

which works out to be 263, 590 ± 10, 924. Since $250,000 falls outside the confidence
interval, Apple Tree can be reasonably confident that their minimum cut-off is met.

5. Statistical interpretation of a confidence interval: Suppose we repeated this sampling

experiment 100 times; that is, we collected 100 different sets of data, each set consisting
of 40 observations. Suppose that we computed a confidence interval based on each of the
100 data sets. On average, we would expect 90 of the confidence intervals to include the
true mean µ; we would expect that 10 would not. The figure 90 comes from the fact that
we chose a 90% confidence interval.

6. More generally, we can choose whatever confidence level we want. The convention is to
specify the confidence level as 1 − α, where α is typically 0.1, 0.05 or 0.01. These three α
values correspond to confidence levels 90%, 95% and 99%. (α is the Greek letter alpha.)

7. Definition: For any α between 0 and 1, we define zα to be the point on the z-axis such
that the area to the right of zα under the standard normal curve is α; i.e.,

P (Z > zα ) = α.

See Figure 1.

8. Examples: If α = .05, then zα = 1.645; if α = .025, the zα = 1.96; if α = .01, then

zα = 2.33. These values are found by looking up 1 − α in the body of the normal table.

9. We can now generalize our earlier 95% confidence interval to any level of confidence 1 − α:
σ
X ± zα/2 √ .
n

3
 0.4 0.4

0.35 0.35

0.3 0.3

0.25 0.25

0.2 0.2

0.15 0.15

α α/2 α/2
0.1 0.1

0.05 0.05

0 0
−4 −3 −2 −1 0 1 2 3 4 −4 −3 −2 −1 0 1 2 3 4
zα −z α/2 z α/2

Figure 1: The area to the right of zα is α. For example, z.05 is 1.645. The area outside ±zα/2
is α/2 + α/2 = α. For example, z.025 = 1.96 so the area to the right of 1.96 is 0.025, the area
to the left of −1.96 is also 0.025, and the area outside ±1.96 is .05.

10. Why zα/2 , rather than zα ? We want to make sure that the total area outside the interval
is α. This means that α/2 should be to the left of the interval and α/2 should be to the
right. In the special case of a 90% confidence interval, α = 0.1, so α/2 = 0.05, and z.05 is
indeed 1.645.
11. Here is a table of commonly used confidence levels and their multipliers:
Confidence Level α α/2 zα/2
90% 0.10 0.05 1.645
95% 0.05 0.025 1.96
99% 0.01 0.005 2.58

Unknown Variance, Large Sample

1. The expression given above for a (1 − α) confidence interval assumes that we know σ 2 , the
population variance. In practice, if we don’t know µ we generally don’t know σ either.
So, this formula is not quite ready for immediate use.
2. When we don’t know σ, we replace it with an estimate. In equation (1) we have an
estimate for the population variance. We get an estimate of the population standard
deviation by taking the square root:


 1
n
s= (Xi − X)2 .
n − 1 i=1
This is the sample standard deviation. A somewhat more convenient formula for
computation is 

 1
n
2
s= ( Xi2 − nX ).
n − 1 i=1

4
 In Excel, this formula is evaluated by the function STDEV.

3. As the sample size n increases, s approaches the true value σ. So, for a large sample size
(n ≥ 30, say),
X −µ
√
s/ n
is approximately normally distributed N (0, 1). Arguing as before, we find that
s
X ± zα/2 √
n

provides an approximate 1 − α confidence level for µ. We say approximate because we

have s rather than σ.

4. Example: Consider the results of the Apple Tree survey given above, but suppose now that
σ is unknown. Suppose the sample standard deviation of the 40 observations collected is
found to be 38900. Then we arrive at the following (approximate) 90% confidence interval
for µ: √ √
(263590 − 1.645(38900/ 40), 263590 − 1.645(38900/ 40)).
The sample mean, the sample size, and the zα/2 are just as before; the only difference is
that we have replaced the true standard deviation with an estimate.

5. Example: Suppose that Apple Tree also wants to estimate the number of customers per
day at nearby stores. Suppose that based on 50 observations they obtain a sample mean
of 876 and a sample standard deviation of 139. Construct a 95% confidence interval for
the mean number of customers per day.

6. Solution: The confidence level is now 95%. This means that α = .05 and α/2 = .025. So
we would look in the normal table to find 1 − .025 = .9750. This corresponds to a z value
of 1.96; i.e., zα/2 = z.025 = 1.96. The rest is easy; our confidence interval is
s √
X ± zα/2 √ = 876 ± (1.96)(139/ 50).
n

7. The expression
σ s
zα/2 √ or zα/2 √
n n
is called the halfwidth of the confidence interval or the margin of error. The halfwidth
is a measure of precision; the tighter the interval, the more precise our estimate. Not
surprisingly, the halfwidth

• decreases as the sample size increases;

• increases as the population standard deviation increases;
• increases as the confidence level increases (higher confidence requires larger zα/2 ).

5
Unknown Variance, Normal Data

1. The confidence interval given above (with σ replaced by s) is based on the approximation

X −µ
√ ≈ N (0, 1).
s/ n

In statistical terminology, we have apprxomiated the sampling distribution of the

ratio on the left by the standard normal. To the extent that we can improve on this
approximation, we can get more reliable confidence intervals.

2. The exact sampling distribution is known when the underlying data X1 , X2 , . . . , Xn comes
from a normal population N (µ, σ 2 ). Recall that in this case, the distribution

X −µ
√ ∼ N (0, 1),
σ/ n

is exact, regardless of the sample size n. However, we are no longer dividing by a con-
stant σ; instead, we are dividing by the random variable s. This changes the distribution.
As a result, the sample statistic
X −µ
t= √
s/ n
is not (exactly) normally distributed; rather, it follows what is known as the Student
t-distribution.

3. The t-distribution looks a bit like the normal distribution, but it has heavier tails, reflect-
ing the added variability that comes from estimating σ rather than knowing it.

4. “Student” refers to the pseudonym of W.S. Gosset, not to you. Gosset identified this
distribution while working on quality control problems at the Guinness brewery in Dublin.
√
5. Once we know the exact sampling distribution of (X − µ)/(s/ n), we can use this knowl-
edge to get exact confidence intervals. To do so, we need to know something about the
t-distribution. It has a parameter called the degrees of freedom and usually denoted by
ν, the Greek letter nu. (This is a parameter in the same way that µ and σ 2 are parameters
of the normal.) Just as Z represents a standard normal random variable, the symbol tν
represents a t random variable with ν degrees of freedom (df).

6. If X1 , . . . , Xn is a random sample from N (µ, σ 2 ), then the random variable

X −µ
t= √
s/ n

has the t-distribution with n − 1 degrees of freedom, tn−1 . Thus, the df is one less than
the sample size.

6
 0.4

0.35

0.3

0.25

0.2

0.15

α
0.1

0.05

0
−4 −3 −2 −1 0 1 2 3 4
t 5,α

Figure 2: The area to the right of tν,α under the tν density is α. The figure shows t5 , the t
distribution with five degrees of freedom. At α = 0.05, t5,.05 = 2.015 whereas z.05 = 1.645. The
tails of this distribution are noticeably heavier than those of the standard normal; see Figure 1.
Right panel shows a picture of “Student” (W.S. Gosset).

7. Just as we defined the cutoff zα by the condition P (Z > zα ) = α, we define a cutoff tν,α
for the t distribution by the condition
P (tν > tν,α ) = α.
In other words, the area to the right of tν,α under the tν density is α. See Figure 2.
Note that tν is a random variable but tν,α is a number (a constant).
8. Values of tν,α are tabulated in a t table. See the last page of these notes. Different rows
correspond to different df; different columns correspond to different α’s.
9. Example: t20,.05 = 1.725; we find this by looking across the row labeled 20 and under
the column labeled 0.05. This tells us that the area to the right of 1.725 under the t20
distribution is 0.05. Similarly, t20,.025 = 2.086 so the area to the right of 2.086 is 0.025.
10. We are now ready to use the t-distribution for a confidence interval. Suppose, once again,
that the underlying data X1 , . . . , Xn is a random sample from a normal population. Then
s
X ± tn−1,α/2 √
n
is a 1 − α confidence interval for the mean. Notice that all we have done is replace zα/2
with tn−1,α/2 .
11. For any sample size n and any confidence level 1 − α, we have tn−1,α/2 > zα/2 . Conse-
quently, intervals based on the t distribution are always wider then those based on the
standard normal.
12. As the sample size increases, the df increases. As the df increases, the t distribution
becomes the normal distribution. For example, we know that z.05 = 1.645. Look down
the .05 column of the t table. tν,.05 approaches 1.645 as ν increases.

7
 13. Strictly speaking, intervals based on the t distribution are valid only if the underlying
population is normal (or at least close to normal). If the underlying population is decidedly
non-normal, the only way we can get a confidence interval is if the sample size is sufficiently
large that X is approximately normal (by the central limit theorem).

14. The main benefit of t-based intervals is that they account for the additional uncertainty
that comes from using the sample standard deviation s in place of the (unknown) popu-
lation standard deviation σ.

15. When to use t, when to use z? Strictly speaking, the conditions are as follows:

• z-based confidence intervals are valid if we have a large sample;

• t-based confidence intervals are valid if we have a sample from a normal distribution
with an unknown variance.

Notice that both of these may apply: we could have a large sample from a normal pop-
ulation with unknown variance. In this case, both are valid and give nearly the same
results. (The t and z multipliers are close when the df is large.) As a practical matter,
t-based confidence intervals are often used even if the underlying population is not known
to be normal. This is conservative in the sense that t-based intervals are always wider
than z-based intervals.

Confidence Intervals for a Proportion

1. Example: To determine whether the location for a new store is a “family” neighborhood,
Apple Tree Supermarkets wants to determine the proportion of apartments in the vicinity
that have 3 bedrooms or more. Field workers randomly sample 100 apartments. Suppose
they find that 26.4% have 3 bedrooms or more. We know that our best guess of the
(unknown) underlying population proportion is the sample proportion p̂ = .264. Can we
supplement this point estimate with a confidence interval?

2. We know that if the true population proportion is p and the sample size is large, then

p(1 − p)
p̂ ≈ N (p, ).
n
As a result,
p̂ − p
 ≈ N (0, 1).
p(1 − p)/n
So a 1 − α confidence interval is provided by

p(1 − p)
p̂ ± zα/2 .
n

8
 
3. The problem, of course, is that we don’t know p, so we don’t know p(1 − p)/n. The
best we can do is replace p by our estimate p̂. This gives us the approximate confidence
interval 
p̂(1 − p̂)
p̂ ± zα/2 .
n
4. Returning to the Apple Tree example, a 95% confidence interval for the proportion is

(.264)(1 − .264)
0.264 ± (1.96) .
100
As before, the 1.96 comes from the fact that P (Z < 1.96) = .975, so P (−1.96 < Z <
1.96) = .95.

Comparing Two Means

1. Often, we are interested in comparing two means, rather than just evaluating them sep-
arately:
• Is the expected return from a managed mutual fund greater than the expected return
from an index fund?
• Are purchases of frozen meals greater among regular television viewers than among
non-viewers?
• Does playing music in the office increase worker productivity?
• Do students with technical backgrounds score higher on statistics exams?
2. In each case, the natural way to address the question is to estimate the two means
in question and determine which is greater. But this is not enough to draw reliable
conclusions. Because of variability in sampling, one sample mean may come out larger
than another even if the corresponding population mean is smaller. To indicate whether
our comparison is reliable, we need a confidence interval for a difference of two
population means.
3. We consider two settings: matched pairs and independent samples. The latter is
easy to understand; it applies whenever the samples collected from the two populations
are independent. The former applies when each observation in one population is paired
to one in the other so that independence fails.
4. Examples of matched pairs:
• If we compare two mutual funds over 30 periods of 6 months each, then the perfor-
mance of the two funds within a single period is generally not independent, because
both funds will be affected by the same world events. If we let X1 , . . . , X30 be the
returns on one fund and Y1 , . . . , Y30 the returns on another, then Xi and Yi are gen-
erally not independent. However, the differences X1 − Y1 , . . . , X30 − Y30 should be
close to independent.

9
 • Consider an experiment in which a software firm plays music in its offices for one
week. Suppose there are 30 programmers in the firm. Let Xi be the number of lines
of code produced by the i-th programmer in a week without music and let Yi be the
number of lines for the same programmer in a week with music. We cannot assume
that Xi and Yi are independent. A programmer who is exceptionally fast without
music will probably continue to be fast with music. We might, however, assume that
the changes X1 − Y1 , . . . , X30 − Y30 are independent; i.e., the effect of music on one
programmer should not depend on the effects on other programmer.

5. Getting a confidence interval for the matched case is easy. Let X1 , . . . , Xn be a random
sample from a population with mean µX and let Y1 , . . . , Yn be matched samples from a
population with mean µY . We want a confidence interval for µX − µY . To get this, we
simply compute the paired differences X1 − Y1 , . . . , Xn − Yn . This gives us a new random
sample of size n. Our point estimator µX − µY is the average difference

1
n
d= (Xi − Yi ) = X − Y .
n i=1

Similarly, the sample standard deviation of the differences is



 1
n
sd =  (Xi − Yi − d)2 .
n − 1 i=1

If the sample size is large, a 1 − α confidence interval for µX − µY is

sd
d ± zα/2 √ .
n
If the underlying populations are normal, then regardless of the sample size an exact
confidence interval is
sd
d ± tn−1,α/2 √ .
n
6. Example: Suppose we monitor 30 computer programmers with and without music. Sup-
pose the sample average number of lines of code produced without music is 195 and that
with music the sample average is 185. Does music affect the programmers? Suppose the
sample standard deviation of the differences sd is 32. Then a 95% confidence interval for
the difference is
32
(195 − 185) ± (1.96) √ = 10 ± 11.45.
30
Since this interval includes zero, we cannot be 95% confident that productivity is affected
by the music.

7. We now turn to the case of comparing independent samples. Let X1 , . . . , Xn1 be a random
sample from a population with mean µX and variance σX2 . Define Y , . . . , Y
1 n2 and µY and
2
σY analogously. We do not assume that the sample sizes n1 and n2 are equal. Let X and
Y be the corresponding sample means.

10
 8. We know that E[X − Y ] = E[X] − E[Y ] = µX − µY . Thus, the difference of the sample
means is an unbiased estimator of the difference of the population means. To supplement
this estimator with a confidence interval, we need the standard error of this estimator.
We know that
V ar[X − Y ] = V ar[X] + V ar[Y ],
by the assumption of independence. This, in turn, is given by
2
σX σ2
+ Y.
n1 n2

So, the standard error of X − Y is


2
σX σ2
+ Y.
n1 n2

9. With this information in place, the confidence interval for the difference follows the usual
pattern. If the sample size is sufficiently large that the normal approximation is valid, a
1 − α confidence interval is

2
σX σ2
(X − Y ) ± zα/2 + Y.
n1 n2

10. Of course, if we don’t know the standard deviations σX and σY , we replace them with
estimates sX and sY .

11. If the underlying populations are normal and we use sample standard deviations, we can
improve the accuracy of our intervals by using the t distribution. The only slightly tricky
step is calculating the degrees of freedom in this setting. The rule is this: Calculate the
quantity
(s2X /n1 + s2Y /n2 )2
.
(sX /n1 )2 /(n1 − 1) + (s2Y /n2 )2 /(n2 − 1)
2

Round it off to the nearest integer n̄. Then n̄ is the appropriate df. The confidence
interval becomes 
s2X s2
(X − Y ) ± tn̄,α/2 + Y.
n1 n2
(We will not cover this case and it will not be on the final exam. Excel does this calculation
automatically; see the Excel help on t-tests.)

Comparing Two Proportions

1. We have seen how to get confidence intervals for the difference of two means. We now
turn to confidence intervals for the difference of two proportions.

11
2. Examples:

• 100 people participate in a study to evaluate a new anti-dandruff shampoo. Fifty

are given the new formula and fifty are given an existing product. Suppose 38 show
improvement with the new formula and only 30 show improvment with the existing
product. How much better is the new formula?
• One vendor delivers 5 shipments out of 33 late. Another vendor delivers 7 out of 52
late. How much more reliable is the second vendor?

3. The general setting is this: Consider two proportions pX and pY , as in the examples
above. Suppose we observe nX outcomes of the first type and nY outcomes of the second
type to get sample proportions p̂X and p̂Y . Our point estimate of the difference pX − pY
is the p̂X − p̂Y . This estimate is unbiased because p̂X and p̂Y are unbiased. The variance
of p̂X − p̂Y is
pX (1 − pX ) pY (1 − pY )
+ ,
nX nY
the sum of the variances. So, the standard error is

pX (1 − pX ) pY (1 − pY )
+ .
nX nY

Since we don’t know pX and pY , we replace then with their estimates p̂X and p̂Y . The
result is the confidence interval

p̂X (1 − p̂X ) p̂Y (1 − p̂Y )
(p̂X − p̂Y ) ± zα/2 + ,
nX nY

valid for large samples.

4. You can easily substitute the corresponding values from the shampoo and vendor examples
to compute confidence intervals for those differences. In the shampoo example, the sample
proportions are 38/50 = .76 and 30/50 = .60. A 95% confidence interval for the difference
is therefore 
.76(1 − .76) .60(1 − .60)
(.76 − .60) ± +
50 50
which is
.16 ± .18.

12
Summary of Confidence Intervals

1. Population mean, large sample:

s
X̄ ± zα/2 √
n

2. Population mean, normal data with unknown variance:

s
X̄ ± tn−1,α/2 √
n

3. Difference of two means, independent samples:


s2X s2
X̄ − Ȳ ± zα/2 + Y
nX nY

4. Difference of two means, matched pairs:

sd
X̄ − Ȳ ± zα/2 √
n
sd
X̄ − Ȳ ± tn−1,α/2 √
n

5. Population proportion, large sample:


p̂(1 − p̂)
p̂ ± zα/2
n

6. Difference of two population proportions, independent large samples:


p̂X (1 − p̂X ) p̂Y (1 − p̂Y )
p̂X − p̂Y ± zα/2 +
nX nY

13
 Cumulative Probabilities for the Standard Normal Distribution. Percentiles of the t Distribution
This table gives probabilities to the left of given z values for the This table gives values with specified probabilities
standard normal distribution to the right of them for the specified df
z 0.00 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 Degrees of Right-hand tail probability alpha
0 0.5000 0.5040 0.5080 0.5120 0.5160 0.5199 0.5239 0.5279 0.5319 0.5359 freedom 0.10 0.05 0.025 0.01
0.1 0.5398 0.5438 0.5478 0.5517 0.5557 0.5596 0.5636 0.5675 0.5714 0.5753 1 3.078 6.314 12.706 31.821
0.2 0.5793 0.5832 0.5871 0.5910 0.5948 0.5987 0.6026 0.6064 0.6103 0.6141 2 1.886 2.920 4.303 6.965
0.3 0.6179 0.6217 0.6255 0.6293 0.6331 0.6368 0.6406 0.6443 0.6480 0.6517 3 1.638 2.353 3.182 4.541
0.4 0.6554 0.6591 0.6628 0.6664 0.6700 0.6736 0.6772 0.6808 0.6844 0.6879 4 1.533 2.132 2.776 3.747
5 1.476 2.015 2.571 3.365
0.5 0.6915 0.6950 0.6985 0.7019 0.7054 0.7088 0.7123 0.7157 0.7190 0.7224
0.6 0.7257 0.7291 0.7324 0.7357 0.7389 0.7422 0.7454 0.7486 0.7517 0.7549 6 1.440 1.943 2.447 3.143
0.7 0.7580 0.7611 0.7642 0.7673 0.7704 0.7734 0.7764 0.7794 0.7823 0.7852 7 1.415 1.895 2.365 2.998
0.8 0.7881 0.7910 0.7939 0.7967 0.7995 0.8023 0.8051 0.8078 0.8106 0.8133 8 1.397 1.860 2.306 2.896
0.9 0.8159 0.8186 0.8212 0.8238 0.8264 0.8289 0.8315 0.8340 0.8365 0.8389 9 1.383 1.833 2.262 2.821
10 1.372 1.812 2.228 2.764
1 0.8413 0.8438 0.8461 0.8485 0.8508 0.8531 0.8554 0.8577 0.8599 0.8621
1.1 0.8643 0.8665 0.8686 0.8708 0.8729 0.8749 0.8770 0.8790 0.8810 0.8830 11 1.363 1.796 2.201 2.718
1.2 0.8849 0.8869 0.8888 0.8907 0.8925 0.8944 0.8962 0.8980 0.8997 0.9015 12 1.356 1.782 2.179 2.681
1.3 0.9032 0.9049 0.9066 0.9082 0.9099 0.9115 0.9131 0.9147 0.9162 0.9177 13 1.350 1.771 2.160 2.650
1.4 0.9192 0.9207 0.9222 0.9236 0.9251 0.9265 0.9279 0.9292 0.9306 0.9319 14 1.345 1.761 2.145 2.624
15 1.341 1.753 2.131 2.602

14
1.5 0.9332 0.9345 0.9357 0.9370 0.9382 0.9394 0.9406 0.9418 0.9429 0.9441
1.6 0.9452 0.9463 0.9474 0.9484 0.9495 0.9505 0.9515 0.9525 0.9535 0.9545 16 1.337 1.746 2.120 2.583
1.7 0.9554 0.9564 0.9573 0.9582 0.9591 0.9599 0.9608 0.9616 0.9625 0.9633 17 1.333 1.740 2.110 2.567
1.8 0.9641 0.9649 0.9656 0.9664 0.9671 0.9678 0.9686 0.9693 0.9699 0.9706 18 1.330 1.734 2.101 2.552
1.9 0.9713 0.9719 0.9726 0.9732 0.9738 0.9744 0.9750 0.9756 0.9761 0.9767 19 1.328 1.729 2.093 2.539
20 1.325 1.725 2.086 2.528
2 0.9772 0.9778 0.9783 0.9788 0.9793 0.9798 0.9803 0.9808 0.9812 0.9817
2.1 0.9821 0.9826 0.9830 0.9834 0.9838 0.9842 0.9846 0.9850 0.9854 0.9857 21 1.323 1.721 2.080 2.518
2.2 0.9861 0.9864 0.9868 0.9871 0.9875 0.9878 0.9881 0.9884 0.9887 0.9890 22 1.321 1.717 2.074 2.508
2.3 0.9893 0.9896 0.9898 0.9901 0.9904 0.9906 0.9909 0.9911 0.9913 0.9916 23 1.319 1.714 2.069 2.500
2.4 0.9918 0.9920 0.9922 0.9925 0.9927 0.9929 0.9931 0.9932 0.9934 0.9936 24 1.318 1.711 2.064 2.492
25 1.316 1.708 2.060 2.485
2.5 0.9938 0.9940 0.9941 0.9943 0.9945 0.9946 0.9948 0.9949 0.9951 0.9952
2.6 0.9953 0.9955 0.9956 0.9957 0.9959 0.9960 0.9961 0.9962 0.9963 0.9964 26 1.315 1.706 2.056 2.479
2.7 0.9965 0.9966 0.9967 0.9968 0.9969 0.9970 0.9971 0.9972 0.9973 0.9974 27 1.314 1.703 2.052 2.473
2.8 0.9974 0.9975 0.9976 0.9977 0.9977 0.9978 0.9979 0.9979 0.9980 0.9981 28 1.313 1.701 2.048 2.467
2.9 0.9981 0.9982 0.9982 0.9983 0.9984 0.9984 0.9985 0.9985 0.9986 0.9986 29 1.311 1.699 2.045 2.462
30 1.310 1.697 2.042 2.457
3 0.9987 0.9987 0.9987 0.9988 0.9988 0.9989 0.9989 0.9989 0.9990 0.9990
3.1 0.9990 0.9991 0.9991 0.9991 0.9992 0.9992 0.9992 0.9992 0.9993 0.9993 40 1.303 1.684 2.021 2.423
3.2 0.9993 0.9993 0.9994 0.9994 0.9994 0.9994 0.9994 0.9995 0.9995 0.9995 60 1.296 1.671 2.000 2.390
3.3 0.9995 0.9995 0.9995 0.9996 0.9996 0.9996 0.9996 0.9996 0.9996 0.9997 120 1.289 1.658 1.980 2.358
3.4 0.9997 0.9997 0.9997 0.9997 0.9997 0.9997 0.9997 0.9997 0.9997 0.9998 z_alpha 1.282 1.645 1.960 2.326