Académique Documents
Professionnel Documents
Culture Documents
The tools and objectives of the Analyze phase are shown in the image below.
Multi-vari Studies
Data can be grouped in terms of sources of variation to help define the way measurements are partitioned. These sources describe characteristics of populations, and the following are common types that describe our everyday processes: Classifications (by category) Geography (of a distribution center or a plant) Geometry (chapters of a book or locations within buildings) People (tenure, job function, or education) Time (deadlines, cycle time, or delivery time)
We can stratify the data to help us understand the way our processes work by categorizing the individual measurements. This helps us understand the variation of the components as it relates to the whole process. For example, errors are being tracked in a process. The variation could be within a subgroup (within a certain batch), between subgroups (from one batch to another batch) or over time (time of day, day of week, shift or even season of the year).
Multi-vari Charts
A multi-vari chart is a tool that graphically displays patterns of variation, and the stability or consistency of a process. It is used to identify possible Xs or families of variation in the preliminary stages of data analysis. Multi-vari charts categorize data as positional (within a subgroup), cyclical (between subgroups), and temporal(over time). You can also detect differences in the inputs (Xs) by using two diagrams presented in the Measure lesson of this course: boxplots (e.g., box-and-whisker plots) when the Y is continuous and the Xs are discrete scatterplots (e.g., scatter diagrams) when the Y is continuous and the Xs are also continuous
Scatterplot Question
Cyclical Variation The number of errors produced (Y) is plotted on the y axis, and the years of experience of operator (X) are plotted on the x axis.
Multi-vari Sampling
Multi-vari analysis uses sources of variation to stratify the data to begin to turn the possible Xs into the probable Xs. Multi-vari analysis helps narrow the focus and serves as a point where we begin to gather samples to determine if the probable Xs are statistically significant. A sampling plan is critical in investigating the largest sources of variation. Sampling plans should be designed in such a way that the resulting data will contain a representative sample of the parameters of interest.
Answer: D
The sample size required to be statistically significant with alpha = 0.05 and a target power of 0.90 = 10. Based on the sample size necessary to achieve the desired level of significance, data were tracked for one week, using check sheets to be completed by mailroom staff, to track the respective variable information by shift.
Additional analysis on the buildings will be conducted to find out why the return mail volumes are different.
One month has passed, and improvements have been implemented. Shown below are boxplots reflecting return mail volume for buildings A and B. Based on these graphs, the variances of building A and B are roughly the same.
Correlation Coefficient
Correlation does not imply causation. Do not assume that a scatter plot pattern means the two variables are related. It is likely that there is a correlation, but possibly not causation. The correlation tool does not establish causation. Relationships over a wider range of data or a different portion of the range of data may exist. For example, there is a strong association between increased income and age or seniority. But you may not automatically assume that your income will increase because you are older, or because you have more experience or seniority.
Table
Change Requests (X) 1 4 5 6 Sum 16 Cost per Transaction (Y) 0.35 0.39 0.40 0.41 1.55 X2 Y2 1 0.1225 0.35 2 16 0.1521 1.56 25 0.16 36 0.1681 2.46 78 0.6027 6.37
(X)(Y)
Calculation:
Also, the residual for each predicted value of Y is defined as the difference between the predicted value and the actual value at X. Assumptions The residuals are normally distributed with a mean of zero and a standard deviation of one. The y-values are independent.
OptionAOptionBOptionC Answer:OptionC
Given a series of values for an independent variable x and the corresponding dependent variable y, we can calculate the estimates of B0 and B1 (which are again the values b0 and b1 respectively). Find the fitted regression line for the following set of data, where x is the independent variable and y is the dependent variable. X Y 3 5 7 9 11 7 9 16 18 23
Calculation/Prediction/Results
Determine the necessary calculations for b1 and b0:
Predict y for a given value of x using the equation for a straight line.
The final fitted regression line for this data set is:
Practice Problem
The customer change request process within the corporate account department is a project of interest for process improvement. The department manager wants to find the relationship between the number of change requests processed per month and the cost per transaction.
X2 1
Y2 .1225
Sum x,y
16 4
1.55 0.3875
78 0.6027 6.37
Calculation: (Solution)
Prediction To predict Y from a given value of X, use the prediction equation: Y-hat = b0 + b1X = 0.3389 + 0.0212X Results Using the equation, we can predict the cost per transaction based on the number of customer change requests. If there are two customer change requests, the cost per transaction is 0.3389 + (0.0121) (2) = 0.363 or approximately $0.36.
In hypothesis testing, the phrase to accept the null hypothesis is not typically used. In statistical terms, the Certified Quality professional can reject the null hypothesis, thus accepting the alternate hypothesis, or fail to reject the null hypothesis. This phrasing is similar to a jury's statement that the defendant is not guilty, which does not say that the defendant is innocent.
General Sequencing
To achieve victory in a project, both practical and statistical improvements are required.
Step 2: Define the practical objective. Define logical categorizations where differences might exist so that meaningful action can be taken. Ask questions to determine what you want to prove (i.e., what questions will the hypothesis test answer?): Are there significant differences in return mail volume between division, product types, and ancillary endorsements? Are there significant differences among the production shifts in variance of system down time? Do error rates vary by shift, product type, and day of the week?
Step 3: Establish hypotheses to answer the practical objective. The following is an example of hypotheses using an Analysis of Variance (ANOVA) test of means where the mean of each shift is equal against the alternative where they are not equal: Null Hypothesis: Ho: 1st shift = 2ndshift = 3rd shift Alternate Hypothesis: Ha: At least one mean is different Step 4: Select the appropriate statistical test. Based on the data that has been collected and the hypothesis test established to answer the practical objective, refer to the Hypothesis Testing Roadmap to select statistical tests. The roadmap is a very important tool to use with each hypothesis test.
Step 5: Define the Alpha () Risk. The Alpha Risk (i.e., Type 1 Error or Producers Risk) is the probability of rejecting the null hypothesis when it is true (i.e., rejecting a good product when it meets the acceptable quality level). Typically, () = 0.05 or 5%. This means there is a 95% (1-) probability you are failing to reject the null hypothesis when it is true (correct decision). For example, in the U.S. legal system, alpha risk is the risk an innocent person could have been convicted of a crime. Step 6: Define the Beta () Risk. The Beta Risk (i.e., Type 2 Error or Consumers Risk) is the probability of failing to reject the null hypothesis when there is significant difference (i.e., a product is passed on as meeting the acceptable quality level when in fact the product is bad). Typically, () = 0.10 or 10%. This means there is a 90% (1-) probability you are rejecting the null hypothesis when it is false (correct decision). Also, the power of the sampling plan is defined as 1 ; hence the smaller the , the larger the power. For example, in the U.S. legal system, beta risk is the risk a guilty person could have been found not guilty and been set free.
Step 7: Establish Delta (). Delta () is the practical significant difference detected in the hypothesis test between:
proportion and a target o Ho: p - po = o Ha: p - po < or Ha: p - po > or Ha: p - po two proportions o Ho: p1 - p2 = o Ha: p1 - p2 < or Ha: p1 - p2 > or Ha: p1 - p2 mean and a target o Ho: - o = o Ha: - o < or Ha: - o > or Ha: - o two means o Ho: 1 - 2 = o Ha: 1 - 2 < or Ha: 1 - 2 > or Ha: 1 - 2
Example
The return mail volume reduction of 10% has been determined to be a practical significance. An error rate difference of 7% has been determined to be of practical significance.
Example
Historically, within the XYZ Company, the average cost of a transaction is $.40 with a standard deviation of $.25. The goal of the project is to reduce the average cost to $.30. Assuming the alpha risk of 5% and a beta risk of 10%, how many samples would be needed? Step 8: Determine the sample size (n). Sample size depends on the statistical test, type of data and alpha and beta risks. There are statistical software packages to calculate sample size. It is also possible to manually calculate sample sizes using a sample size table. Calculating sample size using /: Where delta is the difference in the actual process and the project target or objective divided by the standard deviation
Step 9: Collect the data. Data collection is based on the sampling plan and method.
Step 10: Conduct the statistical tests. Use the Hypothesis Test Roadmap to lead you in the right direction for the type of data you have collected.
Step 11: Develop statistical conclusions. According to Montgomery in Introduction to Statistical Quality Control, the pvalue is the smallest level of significance that would lead to rejection of the null hypothesis (Ho). Typically, if = 0.05 and the p-value 0.05, then reject the null hypothesis and conclude that there is a significant difference. Typically, if = 0.05 and the p-value > 0.05, then fail to reject the null hypothesis and conclude that there is not a significant difference.
Example Ho: 1= 2 = 3 Ha: At least one mean is different With = 0.05 and a computed p-value of 0.00, there is a significant difference between mail endorsement classifications and return mail volume. Therefore, we reject the null hypothesis.
Ho: p1 = p2 = p3 Ha: At least one pi is different With = 0.05 and a computed p-value of 0.12, there is not a significant difference between shift and account processing errors. Therefore, we fail to reject the null hypothesis.
Statistical conclusions can be made by comparing the test statistic (calculated value) to a critical value in a statistical table OR by using a statistical software program to calculate the p-value: The null hypothesis will be rejected if the absolute value of the test statistic is greater than the critical value. For example: o Reject Ho if | tcalc | > tcritical /2, n-2 o Do not reject Ho if | tcalc | < tcritical /2, n-2 If the test statistic (calculated value) = critical value, then it is often considered to be a judgment call. Possible judgment options include repeating the experiment, understanding the real criticality of the answer, etc.
Example Ho: = 0 Ha: 0 Account processing is not on target since tcalc > tcritical (4.74 > 2.262); therefore, reject the null hypothesis.
Ho: 1 = 2 Ha: 1 < 2 Since tcalc < tcritical, fail to reject the null hypothesis: -0.37 > -2.306, we cannot reject the null hypothesis.
Step 12: Determine the practical conclusions. Restate the practical conclusion to reflect the impact in terms of cost, return on investment, technical, etc. Remember, statistical significance does not imply practical significance. Example Statistically the first quarter has the worst error rate and an increased missed account rate of 9%. Through analysis, it has been determined that distribution center G reflects a decreased productivity rate compared to the other distribution centers. In addition to low productivity, there is an increase in employee turnover of 20% for second-shift employees.
The following hypothesis test summary displays the test statistic, statistical tests, and null and alternate hypotheses for means, variances, and proportions.
An example of the One-sample Z-test for population mean, Twosample Z-test for population mean, One-sample T-test, and the Twosample T-test will be shown in this sub-topic. One-way ANOVA and Twoway ANOVA will be discussed later under the specific sub-topic of ANOVA.
Test Statistic
Where: x is the sample mean. can be estimated by s for the sample. n is the sample size.
Reject Ho if:
z < -z /2 or z > z /2
Ha: < 0
z < -z
Ha: > 0
z > z
For Example:
A weight loss company is introducing a new marketing campaign and wishes to include in the advertisement that the average weight loss of a participant is five pounds per month. An associate from the marketing department randomly chooses 75 participant files and finds the sample has an average weight loss of 4.4 pounds per month. Suppose the standard deviation of the population is 0.2. Is the claim of five pounds lost per month valid? The marketing department wants a 95% confidence level in the claim.
Step 1: Establish the hypotheses. Ho: = 5.0 [The claim is 5.0 pounds or more, since this is a weight-loss program.] Ha: < 5.0 (left-tail test) H1 < 5.0 Step 2: Calculate the test statistic.
Step 3: Determine critical value. Use the entire standard normal distribution. The alternative distribution is going to be on the left side of the null distribution, so we put the entire alpha-risk in the left tail of the null distribution, and find the ordinate which cuts off 0.05. On the Standard Normal Distribution Table, this is -1.645, which is an interpolation between -1.64 and -1.65. Step 4: Draw the statistical conclusion. Here are the distributions plotted in actual units, not z-scores. The advertised weight loss has a mean of 5.0 and a standard error of the mean of 0.0239. The actual weight loss has a mean of 4.4 and the same standard error of the mean. The critical region in the null distribution (advertised weight loss) of 0.05 in the left tail is cut off by an ordinate of 4.9620. If the mean of the alternative distribution (actual weight loss) is equal to or greater than 4.9620, then we fail to reject the null hypothesis. However, this is not the case. The mean of the null distribution is 4.4. The probability that it is in the null distribution is less than 0.05 the infamous p-value. If we convert 4.9620 to a z-score, we have (4.9620 5.000) / (0.0239) or -1.645. This is the value of the ordinate that cuts off 0.05 in the left tail of the standard normal distribution.
The steps in completing a two-sample Z-test are the same as the onesample Z-test, but with the appropriate calculation for the test statistic.
Hypotheses
Null hypothesis: Ho: 1= 2 Alternative hypothesis may take one of these forms: Ha: 1 2 Ha: 1 < 2 Ha: 1 > 2
Test Statistic
Where:
x1 and x2 are the sample means. s1 and s2 are the sample standard deviations. n1and n2 are the sample sizes.
Alternative hypothesis (may take one of these forms) Ha: 0 Ha: < 0 Ha: > 0
Test Statistic
Where: x is the sample mean. s is the sample standard deviation. n is the sample size.
Reject Ho if:
t < -t /2, n-1 or t > t /2, n-1
Ha: < 0
Ha: > 0
t > t, n-1
For Example: The Loan Reconciliation Department is reviewing the outstanding general ledger items process to determine if it is on target. The mean balance on an outstanding general ledger item is $5,500 (target value). A sample of 10 outstanding general ledger items was reviewed, reflecting a mean item balance of $5,506 with a standard deviation of $4. Is the process on target? The significance level = 0.05.
Step 1: Establish the hypotheses. Ho: = 5,500 Ha: 5,500 Step 2: Calculate the test statistic.
Step 3: Determine the critical values. The sample size was 10, so df is n -1 or 9. Look at the t.025column (/2) and 9 df. The critical value is 2.262.
Step 4: Draw the statistical conclusion. Reject Ho if tcalc < -t /2 or tcalc > t /2; otherwise, do not reject Ho.
Since tcalc > tcritical, reject the null hypothesis: 4.74 > 2.262 Report back, No, the process is not on target.
Test Statistic
Assumption of equal variance
where:
x1 and x2 are the sample means. s1 and s2 are the sample standard deviations. n1 and n2 are the sample sizes.
Reject Ho if:
t < -t/2, n1+n2-2 or t > t/2, n1+n2-2
Ha: 1 < 2
Ha: 1 > 2
t > t, n1+n2-2
Degrees of Freedom
Assumption of equal variances (n1 + n2) - 2
In a two-sample t-test, degrees of freedom are calculated using the following formula:
Note: The degrees of freedom should be rounded down to be conservative. Look at the t.025 column (/2) and 7 d.f. and you will see the critical values are -2.365 and 2.365.
Two samples
Normal distribution F- Test
For Example
Two check processing centers are being evaluated to see if there is a difference in processing time. We want to know if the processing time variability is significantly different at the two centers. Sample sizes from Center 1 and Center 2 were 10 and 9, respectively, and have sample variances of S12 = 6.89 and S22 = 4.96. Given = 0.05, are the population variances different at a 95% confidence level?
Step 3: Determine the critical values. F distribution tables are entered using the degrees of freedom which are designated by 1 and 2 (nu) where: 1 = n1 - 1 = 10 - 1 = 9 2 = n2 - 1 = 9 - 1 = 8
Look in the table under 9 in the numerator and 8 in the denominator. The right (lower) critical value is given by:
Look in the table under 8 in the numerator and 9 in the denominator. The left (upper) critical value is given by:
Step 4: Draw the statistical conclusion. If Fcalc > Fcritical, then reject the null hypothesis, because the variances are different. Since the calculated F value of 1.39 is between the critical values of 0.31 and3.39, we fail to reject the null hypothesis and cannot conclude that the population variances of Center 1 and Center 2 are different at a 95% confidence level.
Examples of the One-proportion and Two-proportion tests and the steps to calculate them will be shown in this sub-topic. The Chi-square test will be covered in the Contingency Tables and Goodness-of-Fit Test sub-topics later in the course. Situation Summarized data from two or more samples. Sample proportion vs. historic proportion or target. Sample proportion before improvement actions vs. sample proportion after improvement actions. Statistical Test Used Chi-square One-Proportion test Two-Proportion test
One-Proportion Test
The One-proportion test is used when taking a sample from a population, and the number of units of interest is counted in the sample in order to compare the populations mean to some claim. Use When: np and n (1-p) are > 5, Where: n is the sample size p is the proportion
This assumption is necessary for making a normal approximation to the binomial distribution where Z tables can be used. For example, use the One-proportion test to check if there is a statistical difference between: Current fraction defective of a product to the historical fraction defective Current percent of yes votes on a proposition to the percent in a prior election
One-Proportion Test
Null hypothesis Ho: p = po Where po is the claim value for comparing the sample. Alternate hypothesis may take one of these forms: Ha: p po Ha: p < po Ha: p > po
Test Statistic
One-Proportion Test
Critical Values/Reject Regions For Alternative Hypothesis H a: p p o H a: p < p o H a: p > p o z < -z z > z Reject Ho if: z < -z/2 or z > z/2
Two-Proportion Test
The Two-proportion test is used to determine if a proportion from one sample is statistically different than a proportion from another sample. In each sample, units of interest are counted, and a proportion is calculated relative to the total sample in order to compare the proportions to each other. Use when: np and n (1-p) are > 5 where n is the sample size and p is the proportion. This assumption is necessary for making a normal approximation to the binomial distribution where Z tables can be used. For example, use the Two-proportion test to check if there is a statistical difference between: Proportion defective before and after a process improvement Percent accuracy of form completion between two different office locations
Null hypothesis Ho: p1= p2 Where p1 and p2 are the proportions in the populations being compared. Alternative hypothesis may take one of these forms: Ha: p1 p2 Ha: p1 < p2 Test Statistic
Ha: p1 > p2
Step 1: Establish the hypotheses. Ho: p = 0.90 Ha: p < 0.90 Step 2: Calculate the test statistic. Data was collected to calculate p (proportion of accuracy):
p = population proportion n =sample size x = number of items in the sample with the defined attribute p' = sample proportion= x/n = 400/500 = 0.8 p0 = the hypothesized proportion
Step 3: Determine the critical value. Using = 0.05, see the Normal Distribution Table, The test Z value is 1.645.
Step 4: Draw the statistical conclusion. If Zcalc < -Z hypothesis.
critical,
critical,
Step 2 Calculation
Step 3: Determine the critical value. Using = 0.05, see the Normal Distribution Table above. The test Z value is 1.645.
Step 4: Draw the statistical conclusion. If Z calc < -Z critical, then the null hypothesis is rejected. Otherwise, do not reject the null hypothesis.
Since -2.73 < -1.645, we reject the null hypothesis, and conclude that the
Interpreting Results
In testing hypotheses, the null hypothesis is rejected when the test statistic falls in the critical region. These regions were defined by the relationship to the critical values in each of the sample problems. The critical values were based on the level of alpha risk the user was willing to accept. In most computer programs that perform hypothesis testing, the critical values and region are not defined. Instead, the outputs provide a pvaluealong with the test statistic. The probability value (p-value) of a statistical hypothesis test is the probability of getting a value more extreme than the observed value if the null hypothesis were true. The p-value is compared to the desired significance level of our test, and, if it is smaller, the result is significant. Choose appropriate alpha risk, generally 0.05 (95% confidence level). Verify correct sample size. Typically, if p > 0.05, fail to reject the Ho(null) hypothesis. Typically, if p 0.05, reject the Ho(null) hypothesis.
For Example
In this paired t-test for two population means example, each paired sample consists of a member of one population and that members corresponding member in the other population. A candy company will be using a new chocolate recipe, and the plant manager suspects that the cycle time to produce chocolates will be impacted. She is claiming that there will be a need for an increase in staff. Cycle time for a sample size of 10 chocolates was tracked for each type of recipe on a machine that produces the chocolates made from both recipes. Does the data indicate that the new chocolate recipe increases cycle time when = 0.05? Assume that the differences are normally distributed.
Sample # Old recipe (time in secs.) New recipe (time in secs.) Difference, d
1 2 3 4 5 6 7 7 0 6 6 0 8 7 1 9 8 1 5 5 0 5 5 0
7 5 6 -1
8 9 8 8 0 9 7 2
D = 1 - 2 = 0 Ho: D = 0 Ha: D 0 The established alternate hypothesis test is a: Two-tail test when Hais a statement of does not equal () Left-tail test when Ha has the < sign Right-tail test when Ha has the > sign
Step 3: Determine the critical value. Click T-test Table above, to find the critical value using degrees freedom of n -1. Sinced.f. = 10 -1 and = 0.05 with a two-tailed test, the critical value is in the ninth row of the t.025 (2.262).
Step 4: Draw the statistical conclusion. If t calc > t critical or tcalc < -t critical, reject the null hypothesis. Otherwise, do not reject the null hypothesis.
The calculated value of 1.152 is not in the reject region (1.152 < 2.262), so therefore we fail to reject the null hypothesis. At a 95% confidence level, the data suggests that the cycle time did not change using the new chocolate recipe, therefore refuting the claim that additional staff is needed.
Relationship to the Two-sample t-test ANOVA extends the Two-sample t-test for testing the equality of two population means to a more general null hypothesis of comparing the equality of more than two means, versus not all of them being equal. The reason for using ANOVA instead of multiple t-tests is that since the t-test has a chance of being wrong just by sheer luck, the more averages tested, the greater the chance of a false positive. ANOVA corrects that by viewing the whole system at once.
Three call centers (A, B, and C) are being tested to see if customer wait time is statistically the same. A sample of five wait times was taken from each call center. Is there a difference in any of the three call centers? Assume = 0.05.
Step 1: Establish the hypotheses. Ho: 1 = 2 = 3 Ha: At least one of the group means is different from the others. Step 2: Calculate the test statistic. (Step 2 continues on the next page). Calculate the average of each call center (group) and the average of the samples: Call Center A Time 1 Time 2 Time 3 Time 4 Time 5 Average Overall Average 17 23 18 20 19 19.4 20.73 Call Center B 28 26 24 23 26 25.4 Call Center C 16 15 17 18 21 17.4
The sum of squares is calculated by subtracting the average of each call center (group) from the overall average, squaring the result, and adding up the results and multiplying by the number of samples per call center (group): Group Call Center A Call Center B Call Center C Group Average 19.4 25.4 17.4 Overall Average 20.73 20.73 20.73 Difference -1.33 4.67 -3.33 Sum of diff. squared Samples per group SS group Difference2 1.7689 21.8089 11.0889 34.6667 5 34.67 X 5 = 173.4
The error of the sum of squares is calculated by summing the squared differences between each individual and its group mean and then adding the totals to get the sum of squares for the error term: Call Center A 17 23 18 20 19 Group Average 19.4 19.4 19.4 19.4 19.4 Difference -2.4 3.6 -1.4 0.6 -0.4 Difference2 5.76 12.96 1.96 0.36 0.16 Total Call Center A Call Center B 28 Group Average 25.4 Difference 2.6 Difference2 6.76 21.2 SS (Sum of Squares) SS (Sum of Squares)
26 24 23 26
0.36 1.96 5.76 0.36 Total Call Center B 15.2 SS (Sum of Squares)
Call Center C 16 15 17 18 21
21.2 57.6
Step 2: Calculate SSTotal
The total sum of squares is calculated by summing the squared difference of each individual value as compared to the overall average of the data: Call Center A 17 23 18 20 19 Overall Average 20.73 20.73 20.73 20.73 20.73 Difference Difference2 -3.73 2.27 -2.73 -0.73 -1.73 13.91 5.15 7.45 0.53 2.99 Total Call Center A Call Center B 28 26 24 Group Average 20.73 20.73 20.73 Difference Difference 7.27 5.27 3.27 52.85 27.77 10.69
2
SS (Sum of Squares)
23 26
20.73 20.73
2.27 5.27
Call Center C 16 15 17 18 21
Difference Difference -4.73 -5.73 -3.73 -2.73 0.27 22.37 32.83 13.91 7.45 0.07
Total Call Center C Total Sum of Squares Step 2: Calculate the ANOVA table.
76.63 231
Degrees of freedom (df) is calculated for the group, error, and total sum of squares: df group/factor = Number of groups (Call Centers) 1 = 3 1 = 2 df error = (Number of data points 1) (Number of groups 1) = (15 1) (3 1) = 12 df total SS= Number of data points 1 = 15 1 = 14
Complete the ANOVA table to determine the F value. The SSFactor, SSError, SSTotal, and df were all calculated previously. This information is transferred into the following table: Source Group Error Total SS 173.4 57.6 231 df 2 12 14 Mean Square 173.4 / 2 = 86.7 57.6 / 12 = 4.8 F (calc) 86.7 / 4.8 = 18.06
The Mean Square is calculated by dividing the SS by df. F(calc) is determined by dividing the group mean square by the error mean square.
Step 3: Determine the critical value. Fcritical is taken from the F distribution table. From F Distribution Table, for this example , when = 0.05 with df = 2 in the numerator and df = 12 in the denominator, the critical value is 3.89.
Step 4: Draw the statistical conclusion. If Fcalc < Fcritical, fail to reject the null hypothesis. If Fcalc > Fcritical, reject the null hypothesis.
Since Fcalc > Fcritical, we reject the null hypothesis: 18.06 > 3.89. This indicates that at a significance level of 0.05, at least one of the call center average wait times is different.
Goodness-of-Fit Tests
In the goodness-of-fit tests, one is comparing an observed (O) frequency distribution to an expected (E) frequency distribution. The relationship is statistically described by a hypothesis test: Ho: Random variable is distributed as a specific distribution with given parameters. Ha: Random variable does not have the specific distribution with given parameters.
The formula for calculating the Chi-square test statistic for this one-tail test is:
A random sample of size n is taken from the population. The degrees of freedom for this test = k m 1, where: k = number of intervals or cells from the sample to form a frequency distribution m = number of parameters estimated from the sample data
For Example
Management believes that the time between machine breakdowns follows the exponential distribution. We track a bank of identical machines for a number of hours between breakdowns. Test the hypothesis that the distribution is exponential using a 95% ( = 0.05) level of confidence.
Time Between Breakdowns (Interval) Interval Number of Breakdowns 0100 325 >100200 180 >200300 63 >300400 53 >400500 33 >500600 18 >600 Total 6 678
From a sample of 678, the average time to breakdown was estimated at 156.6 hours: = 1/156.6 = 0.0064
Goodness-of-Fit Step 3
Step 3: Calculate the expected values. Interval 0 - 100 >100 -200 >200 - 300 >300 - 400 >400 - 500 >500 - 600 >600 e e e e e e e e
0 -0.0064(100) -0.0064(100) -0.0064(200) -0.0064(300) -0.0064(400) -0.0064(500) -0.0064(600)
Expected Value 320.5 169.0 89.1 47.0 24.7 13.1 14.6 678.0
= 0.0215
= 1.0000
Goodness-of-Fit Step 4
Step 4: Calculate the test statistic. Using the Chi-square test statistic and the values from the previous data tables:
The degrees of freedom is computed as the number of intervals minus the number of parameters estimated (which is 1, as the mean was estimated from the sample), minus 1:
df = 7 1 1 = 5
Since 18.878 > 11.070, we reject the null hypothesis at a 0.05 significance level. The time to breakdown for this equipment is not modeled by the exponential distribution (Ha: Distribution is not exponential).
Exploratory data analysis The analysis is conducted through the use of tools such as: multi-vari studies correlation and regression models ANOVA Chi-square paired comparison tests one-sample tests
Hypothesis testing Another important component of analysis is hypothesis testing. Through hypothesis testing, the Green Belt will define a statement in regard to the significance of certain variables and determine if the variables are related or unrelated. In statistical terms, the Six Sigma Green Belt has two choices: Accept the null hypothesis Reject the null hypothesis
The use of the activities and tools during the Analyze phase of the Six Sigma methodology discovers the root causes by identifying significant Xs impacting process performance. The Green Belt and the team are now ready to move forward in the project generating, selecting, testing, and implementing solutions to address the root causes.