You are on page 1of 46

Confidence Interval for a

Mean
SANDWICH1.XLS
 This file contains the results of a survey done by a
fast food restaurant. The restaurant recently added a
new sandwich to its menu. They conducted a survey
to estimate the popularity of this sandwich.

 A random sample of 40 customers who ordered the


sandwich were surveyed. They were each asked to
rate the sandwich from 1 to 10, 10 being best.

 The manager wants to estimate the mean satisfaction


rating over the entire population of customers by
using a 95% confidence interval. How should she
proceed?
9.2 | 9.3 | 9.4 | 9.5 | 9.6 | 9.7 | 9.8 | 9.9 | 9.10 | 9.11 | 9.12 | 9.13 | 9.14
New Sandwich Data

9.2 | 9.3 | 9.4 | 9.5 | 9.6 | 9.7 | 9.8 | 9.9 | 9.10 | 9.11 | 9.12 | 9.13 | 9.14
The t Distribution
 The t distribution is a close relative of the normal
distribution that appears in a variety of statistical
applications.

 The “degrees of freedom” is a numerical parameter of


the t distribution that defines the precise shape of the
distribution.

 The only difference between the t distribution and the


normal distribution is that it is a little more spread out
and this increase in spread is greater for the small
degrees of freedom.
9.2 | 9.3 | 9.4 | 9.5 | 9.6 | 9.7 | 9.8 | 9.9 | 9.10 | 9.11 | 9.12 | 9.13 | 9.14
TDIST.XLS
 This file contains the sample calculations that
illustrate the TDIST and TINV functions.

 Details to using the TDIST function


– Its first argument must be nonnegative.
– Unlike the NORMDIST function, it returns the probability to
the right of the first argument.
– Its third argument is either 1 or 2 and indicates the number
of tails. By using 1 for this argument, we get the probability in
the right-hand tail only.

9.2 | 9.3 | 9.4 | 9.5 | 9.6 | 9.7 | 9.8 | 9.9 | 9.10 | 9.11 | 9.12 | 9.13 | 9.14
The t Distribution -- continued
 Details of using the TINV function:
– The first argument is the total probability we want in both
tails - half of this goes in the right-hand tail and half goes in
the left-hand tail.
– Unlike the TDIST function, there is no third argument for the
TINV function.
 The t distribution is used when we want to make
inferences about a population mean and the
population standard deviation is unknown.
 Two other close relatives of the normal distribution
are the chi-square and F distribution.
9.2 | 9.3 | 9.4 | 9.5 | 9.6 | 9.7 | 9.8 | 9.9 | 9.10 | 9.11 | 9.12 | 9.13 | 9.14
Confidence Intervals and Levels
 To obtain a confidence interval for the population
mean, we first specify a confidence level, usually
90%, 95%, or 99%.

 We then use the sampling distribution of the point


estimate to determine the multiple of the standard
error we need to go out on either side of the point
estimate to achieve the given confidence level.

 To estimate confidence intervals we use the One-


Sample procedure in Excel’s StatPro add-in.

9.2 | 9.3 | 9.4 | 9.5 | 9.6 | 9.7 | 9.8 | 9.9 | 9.10 | 9.11 | 9.12 | 9.13 | 9.14
Calculation
 The manager must use StatPro’s One-Sample
procedures on the Satisfaction variable.

 To use the procedure place the cursor anywhere in


the data set and select StatPro/Statistical
Inference/One-Sample Analysis menu item.

 In the succeeding dialog boxes, select Satisfaction as


the variable that you want to analyze and then accept
the defaults from there on.

9.2 | 9.3 | 9.4 | 9.5 | 9.6 | 9.7 | 9.8 | 9.9 | 9.10 | 9.11 | 9.12 | 9.13 | 9.14
Results
 The results of the calculation are:
– The best guess for the population mean rating is 6.250, the
sample average in cell F7.

– A 95% confidence interval for the population mean rating


extends from 5.739 to 6.761.

 Thus, the manager can be 95% confident that the


true mean rating over all customers who might try the
sandwich is within this interval.

9.2 | 9.3 | 9.4 | 9.5 | 9.6 | 9.7 | 9.8 | 9.9 | 9.10 | 9.11 | 9.12 | 9.13 | 9.14
Assumptions
 We might question whether the sample is really a
random sample matters.
– The manager may have selected random customers but
more likely she selected 40 consecutive customer. This type
of sample is called a convenience sample. If there isn’t a
reason to assume these 40 differ in any way from the entire
population, then it is safe to assume them as a random
sample.
 Another assumption is that the population distribution
is normal.
– This is probably not a problem because confidence intervals
based on the t distribution are robust to violations of
normality. This means that the resulting confidence intervals
are valid for any populations that are approximately normal.
9.2 | 9.3 | 9.4 | 9.5 | 9.6 | 9.7 | 9.8 | 9.9 | 9.10 | 9.11 | 9.12 | 9.13 | 9.14
Confidence Interval for a
Proportion
Background Information
 The fast food manager from Example 9.1 has already sampled
40 customers to estimate the population mean rating of its new
sandwich.

 Recall that each rating is on a scale of 1 to 10, 10 being best.

 The manager would now like to use the same sample to


estimate the proportion of customer who rate the sandwich at
least 6.

 Her thinking is that these are the customers who are likely to
purchase the sandwich again.

9.2 | 9.3 | 9.4 | 9.5 | 9.6 | 9.7 | 9.8 | 9.9 | 9.10 | 9.11 | 9.12 | 9.13 | 9.14
SANDWICH2.XLS
 This file contains the solution.

 To create this output, we first had to create the


High_rating variable in column C.
– This variable is a 1 for all ratings 6 or larger and 0 is
otherwise.
– We can create this variable with the IF function or the
StatPro/Data Utilities/Create Dummy Variables menu item.

 The confidence interval is formed in rows 14-18 with


formulas and a check of them with the One-Sample
Procedure is in rows 6-11.
9.2 | 9.3 | 9.4 | 9.5 | 9.6 | 9.7 | 9.8 | 9.9 | 9.10 | 9.11 | 9.12 | 9.13 | 9.14
Results

9.2 | 9.3 | 9.4 | 9.5 | 9.6 | 9.7 | 9.8 | 9.9 | 9.10 | 9.11 | 9.12 | 9.13 | 9.14
Results -- continued
 The comparison of the two show slight differences
but these are negligible for large sample sizes.

 The output is fairly good news for the manager.

 Based on this sample of size 40, she can be 95%


confident that the percentage of all customers who
would rate the sandwich 6 or higher is somewhere
between 47.5% and 77.5%.

 This is a large interval so there is a lot of uncertainty.


To reduce this interval the manager would have to
sample more customers.
9.2 | 9.3 | 9.4 | 9.5 | 9.6 | 9.7 | 9.8 | 9.9 | 9.10 | 9.11 | 9.12 | 9.13 | 9.14
Confidence Interval for a
Standard Deviation
Background Information
 A machine produces parts that are supposed to have
diameter 10 centimeters.

 However, due to inherent variability, some diameters


are greater than 10 cm and some are less.

 The production supervisor is concerned with two


things:
– First, he is concerned that the mean diameter might not be
10 centimeters.

– Second, he is worried about the extent of variability in the


diameters.
9.2 | 9.3 | 9.4 | 9.5 | 9.6 | 9.7 | 9.8 | 9.9 | 9.10 | 9.11 | 9.12 | 9.13 | 9.14
Background Information --
continued
 Even if the mean is on target, excessive variability
implies that many of the parts will fail to meet
specifications.

 To analyze the process, he randomly samples 50


parts during the course of a day and measures the
diameter of each part to the nearest millimeter.

9.2 | 9.3 | 9.4 | 9.5 | 9.6 | 9.7 | 9.8 | 9.9 | 9.10 | 9.11 | 9.12 | 9.13 | 9.14
PARTS.XLS
 This file shows the sample results in columns A and
B.

 Should the production supervisor be concerned about


the results from this sample?

9.2 | 9.3 | 9.4 | 9.5 | 9.6 | 9.7 | 9.8 | 9.9 | 9.10 | 9.11 | 9.12 | 9.13 | 9.14
Results

9.2 | 9.3 | 9.4 | 9.5 | 9.6 | 9.7 | 9.8 | 9.9 | 9.10 | 9.11 | 9.12 | 9.13 | 9.14
Calculations
 Because he is concerned about the mean and the
standard deviation of diameters, we obtain 95%
confidence intervals for both.
– This is easy with StatPro’s One-Sample procedure.

– The procedure is the same as before except this time we


check the boxes for both confidence interval options - mean
and standard deviation.

 The confidence interval for the mean extends from


9.986 cm to 10.005 cm. Thus, the supervisor can be
fairly confident that the mean diameter all parts is
close to 10 cm.
9.2 | 9.3 | 9.4 | 9.5 | 9.6 | 9.7 | 9.8 | 9.9 | 9.10 | 9.11 | 9.12 | 9.13 | 9.14
Calculations -- continued
 The confidence interval for the standard deviation
extends from 0.029 cm to 0.043 cm. Is this good or
bad?

 Lets assume the diameter is right on target at 10 cm,


the standard deviation is at the upper limit, 0.043,
and the population distribution is normal. The formula
in cell F26 is
=NORMDIST(10-MacDev,Mean,StDev,1)
+(1-NORMDIST(10+MaxDev,Mean,StDev,1))

 This shows that 13.1% of the parts are unusable.


9.2 | 9.3 | 9.4 | 9.5 | 9.6 | 9.7 | 9.8 | 9.9 | 9.10 | 9.11 | 9.12 | 9.13 | 9.14
Calculations -- continued
 To pursue this analysis further we form a two-way
data table.

 Each value in the data table is the resulting


proportion of unusable parts.

 The best-case scenario (where the mean is close to


the target and the standard deviation is small) still
results in 2.5% unusable parts.

 The message for the supervisor should be clear - he


must work to reduce the underlying variability in the
process.
9.2 | 9.3 | 9.4 | 9.5 | 9.6 | 9.7 | 9.8 | 9.9 | 9.10 | 9.11 | 9.12 | 9.13 | 9.14
Confidence Interval for
the Difference Between
Means
Background Information
 The SureStep Company manufactures high-quality
treadmills for use in exercise clubs.

 SureSteps currently purchases its motors for these


treadmills from supplier A.

 However, it is considering a change to supplier B,


which offers a slightly lower cost. The only question is
whether supplier B’s motors are as reliable as
supplier A’s.

9.2 | 9.3 | 9.4 | 9.5 | 9.6 | 9.7 | 9.8 | 9.9 | 9.10 | 9.11 | 9.12 | 9.13 | 9.14
Background Information --
continued
 To check this SureStep installs motors from supplier
A on 30 of its treadmills and motors from supplier B
on another 30 of its treadmills.

 It then runs these treadmills under typical conditions


and, for each treadmill, records the number of hours
until the motor fails.

 What can SureStep conclude?

9.2 | 9.3 | 9.4 | 9.5 | 9.6 | 9.7 | 9.8 | 9.9 | 9.10 | 9.11 | 9.12 | 9.13 | 9.14
MOTORS.XLS
 The data from the experiment appears in this file.
Here is a portion of that data.

9.2 | 9.3 | 9.4 | 9.5 | 9.6 | 9.7 | 9.8 | 9.9 | 9.10 | 9.11 | 9.12 | 9.13 | 9.14
Boxplots
 In any comparison problem it is a good idea to look
initially at side-by-side boxplots of the two samples.

9.2 | 9.3 | 9.4 | 9.5 | 9.6 | 9.7 | 9.8 | 9.9 | 9.10 | 9.11 | 9.12 | 9.13 | 9.14
Boxplots -- continued
 The boxplots show
– the distribution of times until failure are skewed to the right
for each supplier

– the mean for supplier A is somewhat greater than the mean


for supplier B

– there are several mild outliers

 There seems to be little doubt that supplier A’s motors


will last longer on average than supplier B’s - or is
there?

9.2 | 9.3 | 9.4 | 9.5 | 9.6 | 9.7 | 9.8 | 9.9 | 9.10 | 9.11 | 9.12 | 9.13 | 9.14
Confidence Interval
 A confidence interval for the mean difference allows
us to see whether the differences apparent in the
boxplots can be generalized to all motors from the
two suppliers.

 We find this confidence interval by using StatPro


Two-Sample procedure.

 It shows that the sample means differ by


approximately 93 hours and that the sample standard
deviations are of roughly the same magnitude.

9.2 | 9.3 | 9.4 | 9.5 | 9.6 | 9.7 | 9.8 | 9.9 | 9.10 | 9.11 | 9.12 | 9.13 | 9.14
Confidence Interval -- continued
 The difference between sample means is 93.133
hours, the pooled estimate of the common population
standard deviation is 272.196 hours, the standard
error of the sample mean difference is 70.281 hours;
these values lead to the following 95% confidence
interval for the mean difference: 47.549 to 233.815.

 Not only is this interval wide but it ranges from a


negative value to a positive value.

 If SureStep has to guess they would say that supplier


A’s motors lasted longer, but because of the negative
number there is still a possibility that the opposite is
true.
9.2 | 9.3 | 9.4 | 9.5 | 9.6 | 9.7 | 9.8 | 9.9 | 9.10 | 9.11 | 9.12 | 9.13 | 9.14
Confidence Interval for
the Difference
Between Proportions
Background Information
 An appliance store is about to run a big sale.

 It selects 300 of its best customers and randomly


divides them into two sets of 150 customers each.

 It then mails a notice of the sale to all 300 customers


but includes a coupon for an extra 5% off the sale
price to the second set of customers only.

 As the sale progresses, the store keeps track of


which of these customers purchase appliances.

9.2 | 9.3 | 9.4 | 9.5 | 9.6 | 9.7 | 9.8 | 9.9 | 9.10 | 9.11 | 9.12 | 9.13 | 9.14
COUPONS.XLS
 The data from the sale are recorded in this file.

 What can the store’s manager conclude about the


effectiveness of the coupons?
9.2 | 9.3 | 9.4 | 9.5 | 9.6 | 9.7 | 9.8 | 9.9 | 9.10 | 9.11 | 9.12 | 9.13 | 9.14
Solution
 The data has been arranged in a “contingency” table.

 Of the 150 customers who received coupons, 55


purchased an appliance.

 Of the 150 who did not receive coupons, only 35


purchased an appliance.

 These translate to the sample proportions 0.3667 and


0.2333, calculated in cells B8 and B9 with the formula
=B4/D4 and =B5/D5. By subtraction these lead
directly to the sample difference between proportions,
0.1333, in cell B11.
9.2 | 9.3 | 9.4 | 9.5 | 9.6 | 9.7 | 9.8 | 9.9 | 9.10 | 9.11 | 9.12 | 9.13 | 9.14
Solution -- continued
 The standard error of this difference is calculated in
cell B12 with the formula
=SQRT(SampProp1*(1-SampProp1)/SampSize1
+SampProp2*(1-SampProp2)/SampSize2

 The z-multiple for he confidence interval is calculated


in cell B15 with the formula
=NORMSINV(ConfLev+(1-ConfLev/2)

 Finally, the limits of the confidence interval for the


difference are calculated in cells B18 and B19 with
the formulas
=Diff-zMult*StErr =Diff+zMult*StErr
9.2 | 9.3 | 9.4 | 9.5 | 9.6 | 9.7 | 9.8 | 9.9 | 9.10 | 9.11 | 9.12 | 9.13 | 9.14
Conclusions
 Because the confidence limits are both positive, we
can conclude that the effect of coupons is almost
surely to increase the proportion of buyers.

 How can the store manager interpret this mean


difference? He can use it to estimate the extra
business he will receive by including coupons as
opposed to not including them.

 The confidence interval implies that for every 100


customers, the coupons will probably induce an extra
3 to 23 customers to purchase an appliance who
otherwise would not have made a purchase.
9.2 | 9.3 | 9.4 | 9.5 | 9.6 | 9.7 | 9.8 | 9.9 | 9.10 | 9.11 | 9.12 | 9.13 | 9.14
Conclusions -- continued
 However, the difference between proportions does
not directly indicate the difference in profit from
including coupons.

 This is because the customers with coupons pay 5%


less than the customers without them.

 For every 100 customers who receive a mailing with


no coupon, the store can expect to make $1166.50 in
profit. For every 100 that receive coupons the
expected profit is $1100.10.

 Thus the store makes less money including coupons.


9.2 | 9.3 | 9.4 | 9.5 | 9.6 | 9.7 | 9.8 | 9.9 | 9.10 | 9.11 | 9.12 | 9.13 | 9.14
Controlling Confidence
Interval Length
Background Information
 The fast-food manager in Example 9.1 surveyed 40
customers, each of whom rated a new sandwich on a
scale of 1 to 10.

 Based on the data, a 95% confidence interval for the


mean rating of all potential customers extended from
5.739 to 6.761, for a half-length of (6.761-5.379)/2 =
0.511.

 How large a sample would be needed to reduce this


half-length to approximately 0.3?

9.2 | 9.3 | 9.4 | 9.5 | 9.6 | 9.7 | 9.8 | 9.9 | 9.10 | 9.11 | 9.12 | 9.13 | 9.14
Confidence Intervals
 Confidence intervals are a function of three things:
– the data in the sample
• We have control over the data by using the various random
sampling plans to reduce variability.
• An area of statistics called experimental design suggests how
to perform experiments to obtain the most information from a
given amount of sample data.
– the confidence level
• This effect is clear as the confidence level increases, the length
of the confidence interval increases as well.
– the sample size(s)
• The most obvious way to control confidence interval length is to
choose the size of the sample appropriately.

9.2 | 9.3 | 9.4 | 9.5 | 9.6 | 9.7 | 9.8 | 9.9 | 9.10 | 9.11 | 9.12 | 9.13 | 9.14
Sample Size
 Sample size selection must be done before a sample
is observed.

 Sample size estimation formula:

 1.961.597 
2

n 
 0.3 
2
 z  multiple x est 
n 
 B 

9.2 | 9.3 | 9.4 | 9.5 | 9.6 | 9.7 | 9.8 | 9.9 | 9.10 | 9.11 | 9.12 | 9.13 | 9.14
Calculations
 The formula for n uses three inputs:
– the z multiple, which is 1.96 for a 95% confidence level;

– the prescribed confidence interval half-length B, which is 0.3;

– and an estimate sigmaest of the standard deviation

 The final input must be guessed, but for this example


with a sample size 40 we can use the observed
sample standard deviation of 1.597 (which can be
determined with the STDEV function).

9.2 | 9.3 | 9.4 | 9.5 | 9.6 | 9.7 | 9.8 | 9.9 | 9.10 | 9.11 | 9.12 | 9.13 | 9.14
Calculations -- continued
 The formula yields a rounded result of n = 109.

 The claim, then is that the manager surveys 109


customers with a 95% confidence interval will have
the approximate half length of 0.3.

9.2 | 9.3 | 9.4 | 9.5 | 9.6 | 9.7 | 9.8 | 9.9 | 9.10 | 9.11 | 9.12 | 9.13 | 9.14
Calculations -- continued
 The same calculations can be done using the Sample
Size Selection procedure of Excel’s StatPro add-in.
– Just select the menu item and select the parameter to
analyze and enter the requested values.

– A message telling you the required sample size is displayed.

9.2 | 9.3 | 9.4 | 9.5 | 9.6 | 9.7 | 9.8 | 9.9 | 9.10 | 9.11 | 9.12 | 9.13 | 9.14
Question
 What if the manager was at the planning stage and
didn’t have a “preliminary” sample of size 40? What
standard deviation estimate should she use for est ?

 The manager basically has three choices:


– she can base her estimate of the standard deviation on
historical data assuming that relevant historical data are
available,
– she can take a small preliminary sample (of size 20, say) just
to get an estimate of the standard deviation,
– she can simply guess a value for the standard deviation.
• not recommended but there are some cases where it is the only
feasible option.

9.2 | 9.3 | 9.4 | 9.5 | 9.6 | 9.7 | 9.8 | 9.9 | 9.10 | 9.11 | 9.12 | 9.13 | 9.14

You might also like