In: Statistics and Probability
Problem:
A plant manager is interested in developing a quality-control program for an assembly line that produces light bulbs. To do so, the manager considers the quality of the products that come from the line. The light bulbs are packed in boxes of 12, and the line produces several thousand boxes of bulbs per day. To develop baseline data, workers test all the bulbs in 100 boxes. They obtain the following results:
No. of Defective Bulbs/Box |
No. of Boxes |
0 |
68 |
1 |
27 |
2 |
3 |
3 |
2 |
Run @RISK’s distribution fitting procedure on the preceding data choosing Discrete Sample Data (Counted Format) for the type of data.
a. The Fit Results windows shows that the Poisson is the best fitting theoretical distribution. Is the Poisson a good choice? Why or why not? What is the interpretation of the parameter of the Poisson in this setting?
b. Noticing that there are only two boxes with three defective bulbs, you combine the last two categories in the preceding data. Rerunning @RISK’s fitting procedure, we see that the binomial now fits best according to the Chi-Square measure, with the Poisson coming in a close second. How much has the parameter (m) for the Poisson changed from the case with four categories? Is there a reasonable interpretation of the parameters n and p for the binomial in this setting?
c. Which distribution would you use if you were the plan manager? Why?
a]
Using MINITAB, we test the goodness of fit test for Poisson distribution
Use Goodness-of-Fit Test for Poisson to test the hypotheses:
H0: Data follow a Poisson distribution
H1: Data do not follow a Poisson distribution
Stat > Basic Statistics > Goodness-of-Fit Test for Poisson
In Variable, enter
Click OK.
Session window output
MTB
> PGoodness 'No. of defective';
SUBC> Frequencies 'No. of Boxes';
SUBC> GBar;
SUBC> GChiSQ;
SUBC> Pareto;
SUBC> RTable.
Goodness-of-Fit Test for Poisson Distribution
Data
column: No. of defective
Frequency column: No. of Boxes
Poisson mean for No. of defective = 0.38
No.
of
Poisson
Contribution
defective Observed Probability
Expected to Chi-Sq
0
68 0.677057
67.7057 0.001279
1
27
0.264052
26.4052 0.013398
>=2
5
0.058891
5.8891 0.134229
N DF Chi-Sq
P-Value
100 1
0.148906 0.700
Poisson parameter is 0.38.
Here P-value = 0.7 > . Hence we failed to reject null hypothesis
Conclusion: At , Data follows the Poisson distribution. Poisson is good fit for this data.
b]
If we test Goodness of fit for Binomial test
mean = np = 0.37
Variance = np(1 - p) = 0.33 Calculated from given data after merging last two entries
Parameters of Binomial distribution
p = 1 - Variance/mean = 0.1
n = 0.37/0.1 = 3.7 but n is always integer hence it will be 4
x | P(X =x) | Observed | Expected | (Obs - Exp)^2 | (O - E)^2/E |
0 | 0.6561 | 68 | 65.61 | 5.7121 | 0.087061424 |
1 | 0.2916 | 27 | 29.16 | 4.6656 | 0.16 |
2 | 0.0486 | 5 | 4.86 | 0.0196 | 0.004032922 |
Chi_Sq -Test | 0.25 | ||||
P-value = | 0.12 |
Parameters of Binomial distribution
p = 0.1 and n = 4
Here P-value = 0.12 > . Hence we failed to reject null hypothesis
Conclusion: At , Data follows the Binomial distribution. Binomial is good fit for this data.
c]
If I am plant Manager, I will preferred Poisson distribution because in goodness of fit test of Binomial distribution the sum of observed frequency and sum of Expected frequency is not same. That is Expected frequencies not good.