Você está na página 1de 10

MIDDLE EAST TECHNICAL UNIVERSITY

I E 2 66 – ENGINEERING S T A T I S T I C S II

CASE
STUDY 1

Burcu YÜZÜAK
Cem YOĞURTCU
Onur YILMAZ

Instructor:
YASEMİN SERİN

April 2010, Ankara


Case 1

In the first case it is supposed to make comments about the accuracy on the filling of
first machine.

It is supposed to have a mean of 360, 60 gr of raki and with the weight of bottle it
should be 360. Considering the descriptive statistics output, however, sample mean of
Machine 1 is 357.70 which is not so close to 360.0. Moreover, range of bottles changes from
350.96 to 368.35 and considering the total weight of raki in a bottle, this range appears larger.
Finally standard deviation of 4.12 is quite big for the bottles of 60 gram raki, because
considering a +4.12 dispersion of 15 filling makes a loss of more than whole bottle of raki.

Descriptive Statistics: Machine 1

Variable Percent Mean StDev Variance Minimum Q1 Median Q3


Machine_1 100 357.70 4.12 16.96 350.96 354.41 356.92 360.61

Variable Maximum IQR


Machine_1 368.35 6.20

Observing the histogram and steam-and-leaf display of Machine 1, it is supposed to


have the most frequency around 360. On the other hand, it can be seen that the most frequent
ones located around 353 and 358.

1
Stem-and-Leaf Display: Machine 1

Stem-and-leaf of Machine_1 N = 100


Leaf Unit = 1.0

4 35 0111
20 35 2222223333333333
41 35 444444444455555555555
(16) 35 6666666666677777
43 35 8888888888999999
27 36 000000111111
15 36 2233
11 36 4444455
4 36 67
2 36 88

Finally, considering the lower mean of Machine 1, relatively high standard deviation
and large range of data, it can be said that Machine 1 is not completely accurate.

Case 2

In the second case we are asked to comment on the normality of total raki in packages.

Considering the descriptive statistics output, it can be seen that median and mean of
the distribution is very close. Moreover the small value of skewness shows that the data of
contents of bottles in Table 2 is distributed symmetrically.

Descriptive Statistics: Table 2 - Sadece Raki

Variable Mean StDev Variance Minimum Q1 Median Q3 Maximum


C11 356.13 12.12 146.85 323.93 348.35 356.46 364.60 384.18

Variable Range IQR Skewness


C11 60.25 16.26 0.04

Secondly, looking at histogram of Table 2 data, it is evident that distribution of


frequencies is symmetric around the mean. If we do not take into consideration of some little
changes at tails, it can be easily said that this distribution is completely symmetric.

2
Finally, considering the output of descriptive statistics and histogram of data, it can be
accepted as data of Table 2 is distributed normally.

Case 3

In case three, the company is going to buy a new machine with a claimed standard
deviation of 5 gr. In order to be sure of this claim we will be looking for intervals for standard
deviation.

Analyzing the summary of data related to Table 2 (Only content) gives detailed
information about the confidence intervals. If it is looked at the 95% confidence level for
standard deviation, the interval is (3.803, 7.304) and when it is checked for 99% confidence
level, this range becomes (3.509, 8.332). Moreover, it can be seen that the calculated standard
deviation is 5.001. However, with a further look, it is obvious that ranges of intervals are
different. This difference of ranges shows that if we are going to be more sure about the value
of standard deviation, we will be more open to the different levels of standard deviation. Since
having more information about a variable’s exact value needs gathering more information, in
this concept the company could encounter more distant values of standard deviation, such as
3.509 in 99% confidence level where the manager is sure that 5 gr of standard deviation is
%99 certain.

3
To sum up, although it is hard to reject the claimed value of 5 gr for standard
deviation, being 99% or at least %95 sure of it changes not only the possible outcomes of
standard deviation but also increases the range of these possible outcomes.

4
Case 4

In the forth case, we are asked to ensure that Saadettin will get his delivery with the
mean of at least 60 gr. At the same time our manager says that he can tolerate only 1 % of
time rejection, in order to convince Saadettin and obey our manager, we performed hypothesis
test for the data in Table 2.

According to the hypothesis test output, p-value is zero. It shows that this situation is
not acceptable for Saadettin and we cannot obey our manager at the same time. Because the
smaller the p value, the smaller the probability of that rejecting the hypothesis, which says
that mean equals to 60, is a mistake.

One-Sample Z: Table 2 (Content of bottles)

Test of mu = 60 vs < 60
The assumed standard deviation = 5

99% Upper
Variable N Mean StDev SE Mean Bound Z P
Table 2 (Contents) 20 53.64 5.00 1.12 56.24 -5.69 0.000

Moreover it can be seen from the histogram of Table 2, aimed data which is tested is
out of range, where the desired range is shown with blue line.

5
In the second part, Saadettin also wants to know the probability of getting a delivery
from a population mean content weight 57 gram as if it 60 gram. In this situation, Type II
error occurs. And probability of Type II error is calculated below:

X: weight of the raki’s in bottles

P( n=20

= 1-P( µ=57)

= P(Z = 0.38532

This calculation means that the probability of the time he gets a delivery from a
population with mean content weight 57 gr as if it is 60 gr which means Type ll error is 38, 5
% and it shows that with this probability he could accept our inaccurate delivery.

To sum up, in the first part of the forth case, it is shown that the hypothesis is
absolutely wrong and it is not only inappropriate for Saadettin but also inappropriate for our
manager. In the second part, we showed that he could accept the inaccurate delivery with the
probability of 38.5 %.

Case 5:

In the fifth case, Saadettin wants deliveries consisting of 20 packages and he wants at
most 2 unacceptable packages in a delivery. According to his wishes, an acceptable delivery
should have packages with at least a total of 350 gram of raki and he accepts bottles with the
content of at least 60 gr.

6
In order to figure out whether a delivery is accepted or not, firstly we thought the
acceptance of bottles individually. Using different means under 60 gram such as, 53,64, 50
and 57 we calculated the probability of getting an unacceptable bottle. We assumed that
packages in deliveries are distributed binomially. Using these results in binomial distribution
we calculated the probability of getting an unacceptable delivery.

Test 1:

* 6 = 321.84

Probability of producing an unacceptable package:


X: Weight of raki in bottles

= = = = 0.9892248

Then we assume they distributed binomial; and probability of unacceptable delivery:


Y: number of unacceptable packages in a delivery

Test 2:

*6=300

Probability of producing an unacceptable package:


X: Weight of raki in bottles

= = = = 0.999978

Then we assume they distributed binomial; and probability of unacceptable delivery:


Y: number of unacceptable packages in a delivery

7
Test 3:

*6 = 342

Probability of producing an unacceptable package:


X: Weight of raki in bottles

= = = = 0.743122

Then we assume they distributed binomial; and probability of unacceptable delivery:


Y: number of unacceptable packages in a delivery

= 0,9999999974 1

If we have a further look at these calculations, it is evident that in every trial we get a
probability which is so close 1. Moreover for the supposed means 53,64 and 50 we cannot
calculate the closeness of probability to 1. Finally in the test for 57 gram as mean, we were
able to calculate probability as 0,9999999974.

For these three tests, one for 53,64 which is calculated sample mean from the previous
part, and two more test for nearly ±3 for 53,64, it can be said that with the increase of mean,
the probability of getting an unacceptable delivery is decreasing. This shows that if we were
to increase mean and continue testing, and if we were able to test means larger than 60, we
could have acceptable deliveries according to Saadettin’s demands.

Case 6

In the last case, a new customer, Elçin, wants to buy raki from our company. She is
not concerned about the raki weight in each package but also she does not want to be deceived
and gives a test order. Elçin wants to compare population mean of her test lot with that of
Saadettin and in order to be prepared for the result, we made this comparison by using
hypothesis test with the data given in Table 3.
8
We considered a 2-sample t-test where we are testing the difference between the mean
weights of raki from two lots based on random samples from each. In this case, the null
hypothesis states that the two population means are equal while the alternate hypothesis states
that they are not equal. A p-value below our cutoff level suggests that the population means
are different.

Moreover, we assumed a confidence interval of 99% for this test since Elçin wants a
complete comparison between her and Saadettin’s lots. According to the result of Minitab
output of this test, we could see that the differences of means are 4,02 gram. Considering this
difference with regard to content of a bottle which is 60 gram, it can be said that this
difference is high. Therefore, as soon as Elçin realizes that her sample order has a lower mean
of 4.02 gram in comparison to the Saadettin’s lot, she is going to think that our company is
going to deceive her and she most probably will reject to work with our company.

Two-sample T for Elçin vs Saadettin


N Mean StDev SE Mean
Elçin 50 2155.76 9.50 1.3
Saadettin 150 2159.78 8.86 0.72

Difference = mu (Elçin) - mu (Saadettin)


Estimate for difference: -4.02
99% upper bound for difference: -0.57
T-Test of difference = 0 (vs <): T-Value = -2.73 P-Value = 0.003 DF = 198

To sum up, in the last case according to our comparison with the Elçin’s and
Saadettin’s lots, it can be seen that there is a difference of 4.02 gram in means of these two
lots. Considering her sample with the lower sample mean content of raki, there is no chance
of Elçin accepting to work with us.

Você também pode gostar