Basic Statistical Tests

Click here to start

Basic Statistical Tests

Focus of test Compare groups with each other
Data type

Compare one group with a target

Variables data Attribute data

Click a shape for more information

Data type Continuous data Attribute data

Focus of comparison Center of the distribution (mean or median) Spread of the distribution (standard deviation or variance)

Type of attribute data

Binary data (yes/no)

Count data (occurrences)

Data distribution Normal distribution Nonnormal distribution

Shape of distribution Symmetric 1-Sample Wilcoxon Not symmetric 1-Sample Poisson Rate

1-Sample t

1-Sample Sign

1 Variance

1 Proportion

Click

on any page to return here.

Compare groups with each other

Number of groups Compare two groups

Compare more than two groups

Data type Continuous data Attribute data

How are data sampled Independent samples Paired (dependent) observations

Focus of comparison Spread of distribution (standard deviation or variance) Binary data (yes/no)

Type of attribute data Count data (occurrences)

Center of distribution (mean or median)

Data distribution

Normal distribution 2-Sample t

Nonnormal distribution Mann-Whitney 2 Variances Paired t 2 Proportions 2-Sample Poisson Rate

Click

on any page to return here.

Compare more than two groups

Data type

Continuous data

Attribute data

Focus of comparison

Center of the distribution (mean or median)

Spread of the distribution (standard deviation or variance)

Data distribution

Normal distribution One-Way ANOVA

Nonnormal distribution Kruskal-Wallis Test for Equal Variances Cross Tabulation and Chi-Square

Click

on any page to return here.

Do you want to compare one group with a target or groups with each other?

Compare one group with a target

Compare a characteristic value of a group, such as a mean or standard deviation, with a target value. Often used to evaluate whether a process meets a performance goal. Example A quality analyst wants to determine whether the mean fill weight of a cereal product differs from value stated on the package label (500 g). The analyst weighs a sample of cereal boxes from a single production line.

Compare groups with each other

Compare two or more groups or the same group under different conditions. Often used to evaluate whether a process is affected by different factors, changes, or conditions. Example A sales analyst tracks customer purchases at various store locations. The analyst wants to determine whether the average cost of a customer purchase differs between locations.

Focus of test

Compare one group with a target

Compare groups with each other

To estimate a characteristic value of a group, you must analyze a representative sample of data from the group. Then, you can use hypothesis tests to determine whether that value differs from a target or from another group. A target value may be an industry standard, a past performance level, or a process specification.

Click a shape to move through the decision tree

Click this icon on any page to return to Start.

Basic Statistical Tests

Do you have continuous data or attribute data?

Continuous data
Measures a characteristic of a part or process, such as length, weight, or temperature. The data often includes fractional (or decimal) values.

Attribute data
Counts the presence of a characteristic or condition, such as a physical trait, a type of defect, or a rating, such as pass/fail. The data are whole numbers. Example Managers at a bank want to know whether the proportion of customers who are interested in student loans is high enough (at least 5%) to warrant offering the service to customers. They survey 3500 customers and count how many of them are interested in student loans.

Focus of test: Compare one group with a target

Data type

Example A quality analyst wants to determine whether the mean fill weight differs from the value stated on the package label (500 g). The analyst weighs a sample of cereal boxes from a single production shift.

Continuous data

Attribute data

If possible, collect continuous data because they provide more detailed information. However, sometimes attribute data adequately describe the quality of a part or a process. For example, if you track broken light bulbs, you dont need to measure a characteristic of the bulb to evaluate whether its broken or not. What matters is only the number of bulbs that are broken (counts).

Basic Statistical Tests

Do you want to evaluate the center or the spread of the distribution?

Center of the distribution

Determines whether the central value of the data equals a target value, such as a process specification. Often used to evaluate how accurate a process is. Example An inspector measures a shipment of wheel rims and records their diameters. The inspector evaluates whether the mean diameter of the rims equals the specification diameter of 16 in.

Spread of the distribution

Determines whether the variation of the data equals a target value, such as an industry standard. Often used to evaluate how consistent a process is. Example The manager of a lumberyard wants to analyze how consistently a sawmill produces beams that are on average 100 cm long. The manager measures the beams and analyzes whether the standard deviation of the lengths is less than 1 mm.

Focus of test: Compare one group with a target Data type: Continuous

Focus of comparison

Center of the distribution

Spread of the distribution

You can evaluate the central value of the data using the mean (the arithmetic average) or the median (the middle value when the data are arranged from lowest to highest). The median is less sensitive than the mean to outliers. You can evaluate the variation, or spread, of the data using the standard deviation or the variance. The standard deviation is often easier to interpret because it uses the same units as the data.

Basic Statistical Tests

Does your data follow a normal or nonnormal distribution?

Normal distribution
Data that follow a symmetric, bell-shaped distribution. Example An auto parts company manufactures wheel rims to be 16 inches in diameter. Inspectors randomly sample wheel rims and measure their diameters. They determine that the measurements follow a normal distribution.

Nonnormal distribution
Data that follow a distribution that either is not symmetric or is symmetric but not bellshaped. Example A shipping company guarantees delivery of domestic packages within 72 hours. Inspectors randomly sample orders and record the delivery times. They determine that the times are not symmetrically distributed and, therefore, are not normal.

Focus of test: Compare one group with a target Data type: Continuous Focus of comparison: Center of the distribution

Data distribution

Normal distribution

Nonnormal distribution

The distribution of your data often depends on your process. For example, data that track cycle time for service processes, such as the time needed to process an application, serve a customer, or deliver a product, often do not follow a normal distribution. Although the normal distribution is always bell-shaped, not all bell-shaped distributions are normal. To determine whether your data follow a normal distribution, use normal probability plots, histograms, or normality tests such as the Anderson-Darling test. You can perform these evaluations using Minitabs Normality Test or Graphical Summary. If you have 30 or more observations, you can generally treat your data as being normal and perform a basic statistical test to compare the mean of the group with a target value.

Basic Statistical Tests

Basic Statistical Tests

1-Sample t

1-Sample t
The 1-sample t-test determines whether the mean differs from a target value and provides a range of values that is likely to include the true mean. Example An inspector samples a shipment of wheel rims and measures their diameters. The inspector uses a 1-sample t-test to determine whether the mean diameter of the rims differs from the specification of 16 in. To perform a 1-sample t-test in Minitab, choose Stat > Basic Statistics > 1-Sample t.
One-Sample T: Diameter

Focus of test: Compare one group with a target Data type: Continuous Focus of comparison: Center of the distribution Data distribution: Normal

Test of mu = 16 vs not = 16 Variable Diameter N 20 Mean 15.9992 StDev 0.0034 SE Mean 0.0008 95% CI (15.9976, 16.0007) T -1.12 P 0.278

1-Sample t

Use Minitabs 1-sample t-test to evaluate whether the mean is greater than, less than, or not equal to a target value. If you choose a one-sided test (less than or greater than), the test has more power to detect a difference in the direction of interest. However, it cannot detect a difference in the opposite direction. For example, you want to know whether the mean diameter of the wheel rims differs from 16 in. You could increase the power of the test by testing only whether the mean is greater than 16 in. However, this test does not detect a difference if the mean is less than 16 in. Typically, you use a 1-sample t-test to evaluate continuous data that are normally distributed. You can also use a 1-sample t-test to evaluate data that is nonnormal but meets one of these conditions: has 30 or more observations, is symmetric with at least 5 observations, or has relatively few extreme outliers.To quickly evaluate normality and outliers, use Minitabs Graphical Summary. If your data contains any extreme outliers, make sure they are valid measurements. Use Power and Sample Size for 1-sample t to determine how much data you need to detect an important difference between the mean and the target value.

Basic Statistical Tests

Is the data symmetric or not symmetric?

Symmetric
The data are distributed evenly on both sides of the central value. Example A pharmaceutical company wants to determine whether a newly developed drug relieves symptoms in less than 12 minutes. A researcher administers the drug to 16 patients and records the time elapsed until symptoms abate. Although the data are not normal, they are symmetrically distributed.

Not symmetric
The data are not distributed evenly on both sides of the central value and are skewed in one direction. Example A stainless steel manufacturer wants to determine whether the percentage of chromium in its products equals 18%. An inspector measures the chromium content in a sample of products and determines that the data do not follow a symmetric distribution.

Focus of test: Compare one group with a target Data type: Continuous Focus of comparison: Center of the distribution Data distribution: Nonnormal

Shape of distribution

Symmetric

Not symmetric

To evaluate symmetry, display your data in a histogram or boxplot to determine whether it is similarly distributed on both sides of the central data value. If the data are symmetric, the mean and the median are approximately equal, so you can use either the mean or the median to describe the center of the data. You can also use Minitabs Symmetry Plot to evaluate the symmetry of your data. Choose Stat > Quality Tools > Symmetry Plot.

Basic Statistical Tests

Basic Statistical Tests

1-Sample Wilcoxon

1-Sample Wilcoxon
The 1-sample Wilcoxon test determines whether the median differs from a target value and provides a range of values that is likely to include the true median. Example A researcher at a pharmaceutical company administers a newly developed drug to 16 patients and records the time elapsed until symptoms abate. The data do not follow a normal distribution but are symmetric. The researcher uses a 1-sample Wilcoxon test to determine whether the drug relieves symptoms in less than 12 minutes. To perform a 1-sample Wilcoxon test in Minitab, choose Stat > Nonparametrics > 1-Sample Wilcoxon.
Wilcoxon Signed Rank Test: Time Test of median = 12.00 versus median < 12.00 N for Test 15 Wilcoxon Statistic 19.0 Estimated Median 9.350

Focus of test: Compare one group with a target Data type: Continuous Focus of comparison: Center of the distribution Data distribution: Nonnormal Shape of distribution: Symmetric

Time

N 15

P 0.011

Wilcoxon Signed Rank CI: Time Confidence Achieved Interval Confidence Lower Upper 95.0 7.30 11.40

1-Sample Wilcoxon
Time N 15

Estimated Median 9.35

The median is the middle value of the data when they are arranged from lowest to highest. The median is less sensitive to outliers than the mean. You can determine whether the median is greater than, less than, or not equal to a target value. If you choose a one-sided test (less than or greater than), the test has more power to detect a difference in the direction of interest. However, it cannot detect a difference in the opposite direction. You can also use a 1-sample t-test to evaluate data that is nonnormal but meets one of these conditions: has 30 or more observations, is symmetric with at least 5 observations, or has relatively few extreme outliers. To quickly evaluate normality and outliers, use Minitabs Graphical Summary. If your data contains any extreme outliers, make sure they are valid measurements.

Basic Statistical Tests

Basic Statistical Tests

1-Sample Sign

1-Sample Sign
The 1-sample sign test determines whether the median differs from a target value and provides a range of values that is likely to include the true median. Example An inspector for a stainless steel manufacturer measures the chromium content in 12 product samples and determines that the data do not follow a symmetric distribution. The inspector uses a 1-sample sign test to determine whether the median chromium content differs from the specification of 18%. To perform a 1-sample sign test in Minitab, choose Stat > Nonparametrics > 1-Sample Sign.
Sign Test for Median: %Chromium Sign test of median = N 12 Below 8 18.00 versus not = 18.00 Equal 0 Above 4 P 0.3877 Median 17.70

Focus of test: Compare one group with a target Data type: Continuous Focus of comparison: Center of the distribution Data distribution: Nonnormal Shape of distribution: Not symmetric

%Chromium

Sign CI: %Chromium Sign confidence interval for median Confidence Achieved Interval Confidence Lower Upper Position 0.8540 17.50 18.10 4 0.9500 17.43 18.76 NLI 0.9614 17.40 19.00 3

1-Sample Sign
%Chromium

N 12

Median 17.70

The median is the middle value of the data when they are arranged from lowest to highest. The median is less sensitive to outliers than the mean and can be a more informative measure of the center when the data do not follow a symmetric distribution. You can evaluate whether the median is greater than, less than, or not equal to a target value. If you choose a one-sided test (less than or greater than), the test has more power to detect a difference in the direction of interest. However, it cannot detect a difference in the opposite direction. You can also use a 1-sample t-test to evaluate data that is nonnormal but meets one of these conditions: has 30 or more observations, is symmetric with at least 5 observations, or has relatively few extreme outliers. To quickly evaluate normality and outliers, use Minitabs Graphical Summary. If your data contains any extreme outliers, make sure they are valid measurements.

Basic Statistical Tests

Basic Statistical Tests

1 Variance

1 Variance
The 1 variance test determines whether the variation (standard deviation or variance) differs from a target value and provides a range of values that is likely to include the true variation. Example A manager of a lumberyard measures the length of 50 beams cut by a sawmill. The manager uses a 1 variance test to evaluate whether the standard deviation of the beam lengths is less than 1 mm. To perform a 1 variance test in Minitab, choose Stat > Basic Statistics > 1 Variance.
Test and CI for One Standard Deviation: Length Method Null hypothesis Alternative hypothesis Sigma = 1 Sigma not = 1

Focus of test: Compare one group with a target Data type: Continuous Focus of comparison: Spread of the distribution

1 Variance

The standard method is only for the normal distribution. The adjusted method is for any continuous distribution. Statistics Variable Length N 50 StDev 0.871 Variance 0.759

95% Confidence Intervals Variable Length Tests Variable Length Method Standard Adjusted Chi-Square 37.17 28.89 DF 49.00 38.09 P-Value 0.215 0.282 Method Standard Adjusted CI for StDev (0.728, 1.085) (0.712, 1.122) CI for Variance (0.529, 1.178) (0.507, 1.259)

You can use Minitabs 1 Variance test on normal or nonnormal continuous data. You can evaluate whether the standard deviation (or variance) is greater than, less than, or not equal to a target value. If you choose a one-sided test (less than or greater than), the test has more power to detect a difference in the direction of interest. However, it cannot detect a difference in the opposite direction.

Basic Statistical Tests

Do you have binary data or count data?

Binary data
Classifies items into one of two categories, such as pass/fail or yes/no. Often used to compare a proportion.

Count data
Counts the presence of a characteristic, result, or activity. Often used to compare an occurrence rate. Example Inspectors at a bus company count the number of bus breakdowns each day for 30 days. The company wants to determine the daily rate of bus breakdowns.

Focus of test: Compare one group with a target Data type: Attribute

Type of attribute data

Example A direct-mail firm wants to track the proportion of customers who respond to a direct-mail advertisement by purchasing the advertised item. Marketing analysts randomly sample 1000 customers who received the mailer and record whether or not the customer bought the item.

Binary data

Count data

For count data, you count the number of occurrences in a given amount of time, area, volume, or other observation space. If a process has a constant rate of occurrence, use data from any observation space to make an inference about the entire process, such as the number of defects per day, per month, or per year. Otherwise, make sure that the observation space over which you collect the data is appropriate for the question you want to answer. Suppose the rate of phone calls at a call center varies greatly each hour. If you track only the calls received between 8 am and 9 am, the rate will not reflect the average hourly rate over the entire day. Conversely, if you track the calls over the entire day, the average hourly rate will not reflect the rate of calls received between 8 am and 9 am.

Basic Statistical Tests

Basic Statistical Tests

1 Proportion

1 Proportion
A 1 proportion test determines whether a proportion differs from a target value and provides a range of values that is likely to include the true proportion. Example A marketing analyst at a direct-mail firm randomly samples 1000 customers who received an advertising mailer and records whether or not the customer bought the item. The analyst uses a 1 proportion test to determine whether the proportion of customers who respond to the mailer differs from the national average of 6.5%. To perform a 1 proportion test in Minitab, choose Stat > Basic Statistics > 1 Proportion.
Test and CI for One Proportion Test of p = 0.065 vs p not = 0.065 Sample 1 X 87 N 1000 Sample p 0.087000 95% CI (0.070268, 0.106208) Exact P-Value 0.008

Focus of test: Compare one group with a target Data type: Attribute Type of attribute data: Binary

1 Proportion

You can evaluate whether the proportion is greater than, less than, or not equal to the target value. If you choose a one-sided test (less than or greater than), the test has more power to detect a difference in the direction of interest. However, it cannot detect a difference in the opposite direction. Suppose you want to know whether the proportion of customers who respond to the mailer differs from 6.5%. You could increase the power of the test by testing only whether the proportion is greater than 6.5%. However, this test does not detect a difference if the proportion is less than 6.5%. Use Power and Sample Size for 1 proportion to determine how much data you need to detect an important difference between the proportion and the target value.

Basic Statistical Tests

Basic Statistical Tests

1-Sample Poisson Rate

1-Sample Poisson Rate

A 1-sample Poisson rate determines whether a rate differs from a target value and provides a range of values that is likely to include the true rate. Example Inspectors at a bus company count the number of bus breakdowns each day for 30 days. The inspectors use a 1-sample Poisson rate to determine the daily rate of bus breakdowns and see whether it differs from the daily breakdown rate from the previous year (2.1 buses/day). To perform a 1-sample Poisson rate in Minitab, choose Stat > Basic Statistics > 1-Sample Poisson Rate.
Test and CI for One-Sample Poisson Rate Test of rate = 2.1 vs rate not = 2.1 Sample 1 Total Occurrences 94 N 30 Rate of Occurrence 3.13333 95% CI (2.53205, 3.83441) Exact P-Value 0.000

Focus of test: Compare one group with a target Data type: Attribute Type of attribute data: Count

1-Sample Poisson Rate

Length of observation = 1.

You can evaluate whether the rate is greater than, less than, or not equal to the target value. If you choose a one-sided test (less than or greater than), the test has more power to detect a difference in the direction of interest. However, it cannot detect a difference in the opposite direction. Suppose you want to know whether the rate of bus breakdowns differs from 2.1. You could increase the power of the test by testing only whether the daily rate of bus breakdowns is less than 2.1 per day. However, this test does not detect a difference if the rate is greater than 2.1 per day.

Basic Statistical Tests

How many groups are you comparing?

Two groups
Compare two groups or the same group under two conditions. Example Hospital researchers want to compare postsurgery recovery times in two groups of patients: patients who receive follow-up care at the hospital and patients who receive followup care at home.

More than two groups

Compare more than two groups or the same group under more than two conditions. Example A quality engineer wants to compare the mean strength of nylon fiber from 5 different suppliers to see whether the strength of the nylon differs between the suppliers.

Focus of test: Compare groups with each other

Number of groups

Two groups

More than two groups

Basic Statistical Tests

Do you have continuous data or attribute data?

Continuous data
Measures a characteristic of a part or process, such as length, weight, or temperature. The data often includes fractional (or decimal) values.

Attribute data
Counts the presence of a characteristic or condition, such as a physical trait, a type of defect, or a rating, such as pass/fail. The data are whole numbers. Example An analyst wants to compare the quality of light bulbs from two different suppliers. The analyst inspects a sample of bulbs from each supplier and counts the number of broken bulbs in each sample.

Focus of test: Compare groups with each other Number of groups: Two

Example A quality engineer wants to evaluate whether the mean weight of cereal boxes is the same across two production shifts. The analyst samples boxes from each shift and records their weights.

Data type

Continuous data

Attribute data

If possible, collect continuous data because they provide more detailed information. However, sometimes attribute data adequately describe the quality of a part or a process. For example, if you track broken light bulbs, you dont need to measure a characteristic of the bulb to evaluate whether its broken or not. What matters is only the number of bulbs that are broken (counts).

Basic Statistical Tests

Do you have independent samples or paired (dependent) observations?

Independent samples
Each sample is selected randomly, so the observations in one sample do not affect the observations in the other sample. Example A health management company wants to compare satisfaction ratings from former patients at two hospitals. Because the two patient groups are composed of different individuals who are randomly selected, the samples are independent.

Paired observations
The sample is a set of matched pairs, so that one observation is dependent on (related to) the other observation in the pair. Example To evaluate a computer training course, a human resource analyst tests the computer skills of each trainee before and after the training. Therefore, each post-training test score is associated (paired) with a pre-training test score from the same trainee.

Focus of test: Compare groups with each other Number of groups: Two

Data type: Continuous

How are data sampled

Independent samples

Paired observations

Independent samples occur when you: Randomly sample items or subjects from two groups (populations). For example, a quality engineer takes a random sample of bolts that are produced by different machines. The bolts from each machine are considered independent samples. Randomly split a sample into groups and apply distinct conditions to each group. For example, an analyst randomly assigns a sample of customers to two groups and sends each group a different promotional email. The analyst records whether each customer responds to the email. Paired (dependent) observations occur when you: Measure a characteristic of the same subject under different conditions. For example, a researcher measures cholesterol in the same patients before and after they follow a low-fat diet. Apply the same conditions to two items or subjects. For example, a quality engineer measures the tread wear in two brands of tires that were put on the same car. The tread wear for the two tires is likely to depend on the car.
Basic Statistical Tests

Do you want to compare the center or the spread of each distribution?

Center of distribution
Compares the central value in the measurements for each group. You want to determine whether the groups differ. Example A quality analyst at a health management company records satisfaction ratings from former patients at two hospitals. The analyst wants to compare the average satisfaction ratings at each hospital to determine whether they differ.

Spread of distribution
Compares the variation in the measurements for each group. You want to determine whether the data vary more in one group than in the other. Example A manager of a lumberyard measures the length of beams that are cut by two different sawmills. The manager wants to compare the spread of the measurements for each sawmill to determine whether the consistency of the beam lengths is the same.

Focus of test: Compare groups with each other Number of groups: Two

Data type: Continuous

How are data sampled: Independent samples

Focus of comparison

Center of distribution

Spread of distribution

You can evaluate the center of the data using the mean (the arithmetic average) or the median (the middle value when the data are arranged from lowest to highest). The median is less sensitive than the mean to outliers. You can evaluate the variation, or spread, of the data using the standard deviation or the variance. The standard deviation is often easier to interpret because it uses the same units as the data.

Basic Statistical Tests

Does your data follow a normal or nonnormal distribution?

Normal distribution
Data that follow a symmetric, bell-shaped distribution. Example A quality analyst at a health management company records the patient satisfaction ratings at two hospitals. The analyst determines that the satisfaction ratings for each hospital follow a normal distribution.

Nonnormal distribution
Data that follow a distribution that either is not symmetric or is symmetric but not bellshaped. Example A public transportation department uses two brands of paint for road stripes. For each paint brand, an analyst records the number of months that stripes last on the highway. The analyst determines that the durations for each paint brand are not symmetrically distributed and, therefore, are not normal.

Focus of test: Compare groups with each other Number of groups: Two

Data type: Continuous

How are data sampled: Independent samples

Focus of comparison: Center of distribution

Data distribution

Normal distribution

Nonnormal distribution

The distribution of your data often depends on your process. For example, data that track cycle time for service processes, such as the time needed to process an application, serve a customer, or deliver a product, often do not follow a normal distribution. Although the normal distribution is always bell-shaped, not all bell-shaped distributions are normal. To determine whether your data follow a normal distribution, use normal probability plots, histograms, or normality tests such as the Anderson-Darling test. You can perform these evaluations using Minitabs Normality Test or Graphical Summary. If you have 30 or more observations in each sample, you can generally treat your data as being normal and perform a 2-sample t-test to compare the means of two groups.

Basic Statistical Tests

Basic Statistical Tests

2-Sample t

2-Sample t
The 2-sample t-test determines whether the means of two groups differ and provides a range of values that is likely to include the true difference between the means. Example A quality analyst at the health management company wants to compare patient satisfaction ratings at two hospitals. The analyst uses a 2-sample t-test to evaluate whether the average satisfaction ratings from former patients differ between hospitals. To perform a 2-sample t-test in Minitab, choose Stat > Basic Statistics > 2-Sample t.
Two-Sample T-Test and CI: A, B

Focus of test: Compare groups with each other Number of groups: Two

Data type: Continuous

Two-sample T for A vs B A B N 10 15 Mean 80.7 59.0 StDev 10.6 14.2 SE Mean 3.4 3.7

How are data sampled: Independent samples

Focus of comparison: Center of distribution

Difference = mu (A) - mu (B) Estimate for difference: 21.70 95% CI for difference: (10.79, 32.61) T-Test of difference = 0 (vs not =): T-Value = 4.11 Both use Pooled StDev = 12.9213

P-Value = 0.000

DF = 23

Data distribution: Normal

2-Sample t

Use Minitabs 2-sample t-test to evaluate whether the mean of one group is greater than, less than, or not equal to the mean of the other group. If you choose a one-sided test (less than or greater than), the test has more power to detect a difference in the direction of interest. However, it cannot detect a difference in the opposite direction. Typically, you use a 2-sample t-test to evaluate continuous data that are normally distributed. You can also use a 2-sample t-test to evaluate data that is nonnormal but meets one of these conditions: has 30 or more observations, is symmetric with at least 5 observations, or has relatively few extreme outliers.To quickly evaluate normality and outliers, use Minitabs Graphical Summary. If your data contains any extreme outliers, make sure they are valid measurements. Use Power and Sample Size for 2-sample t to determine how much data you need to detect an important difference between the means of the two groups.

Basic Statistical Tests

Basic Statistical Tests

Mann-Whitney

Mann-Whitney
The Mann-Whitney test determines whether the medians of two groups differ and provides a range of values that is likely to include the true difference between the medians. Example A public transportation department uses two brands of paint for road stripes. An analyst records the number of months that each brand of paint lasts on the highway. Because the samples are relatively small and the data is not normal, the analyst uses a Mann-Whitney to determine whether the duration of the two paints differs. To perform a Mann-Whitney test in Minitab, choose Stat > Nonparametrics > Mann-Whitney.
Mann-Whitney Test and CI: Brand A, Brand B

Focus of test: Compare groups with each other Number of groups: Two

Data type: Continuous

Brand A Brand B

N 11 10

Median 36.00 37.50

How are data sampled: Independent samples

Focus of comparison: Center of distribution

Point estimate for ETA1-ETA2 is -1.60 95.5 Percent CI for ETA1-ETA2 is (-3.30,0.20) W = 94.5 Test of ETA1 = ETA2 vs ETA1 not = ETA2 is significant at 0.0671 The test is significant at 0.0668 (adjusted for ties)

Data distribution: Nonnormal

The median is the middle value of the data when they are arranged from lowest to highest. The median is less sensitive to outliers than the mean and can be be a more informative measure of the center when the data do not follow a symmetric distribution. You can evaluate whether the median of one group is greater than, less than, or not equal to the median of the other group. If you choose a one-sided test (less than or greater than), the test has more power to detect a difference in the direction of interest. However, it cannot detect a difference in the opposite direction. For the Mann-Whitney test, both groups should have approximately the same variance. To compare the variances of the two groups, use Minitabs 2 Variances. You can also use the 2-sample t-test to evaluate data that is nonnormal but meets one of these conditions: has 30 or more observations, is symmetric with at least 5 observations, or has relatively few extreme outliers.To quickly evaluate normality and outliers, use Minitabs Graphical Summary. If your data contains any extreme outliers, make sure they are valid measurements.

Mann-Whitney

Basic Statistical Tests

Basic Statistical Tests

2 Variances

2 Variances
The 2 variances test determines whether the variances (or standard deviations) of two groups differ and provides a range of values that is likely to include the true variation in each group. Example A manager of a lumberyard wants to compare the consistency of the length of beams cut by two different sawmills. The manager uses a 2 variances test to determine whether the variation of the beam lengths differs between the sawmills. To perform a 2 variances test in Minitab, choose Stat > Basic Statistics > 2 Variances.

Focus of test: Compare groups with each other Number of groups: Two

Data type: Continuous

How are data sampled: Independent samples

Focus of comparison: Spread of distribution

2 Variances

You can perform the 2 variances test on normal or nonnormal data.

Basic Statistical Tests

Basic Statistical Tests

Paired t

Paired t
The paired t-test examines the mean difference between paired observations and provides a range of likely values for the difference. Example A human resource analyst evaluates the computer skills of each trainee before and after training. The analyst uses a paired t-test to determine whether the mean test scores before and after the training differ. To perform a paired t-test in Minitab, choose Stat > Basic Statistics > Paired t.
Paired T-Test and CI: Before, After Paired T for Before - After

Focus of test: Compare groups with each other Number of groups: Two

Data type: Continuous

Before After Difference

N 15 15 15

Mean 73.20 83.47 -10.27

StDev 15.73 10.23 8.52

SE Mean 4.06 2.64 2.20

How are data sampled: Paired observations

95% CI for mean difference: (-14.99, -5.55) T-Test of mean difference = 0 (vs not = 0): T-Value = -4.67

P-Value = 0.000

Paired t

Typically, you use a paired t-test to evaluate the mean differences between paired observations when the differences follow a normal distribution. You can also use a paired t-test to evaluate paired data that is nonnormal but meets one of these conditions: has 30 or more observations, is symmetric with at least 5 observations, or has relatively few extreme outliers.To quickly evaluate normality and outliers, use Minitabs Graphical Summary. If your data contains any extreme outliers, make sure they are valid measurements. If the paired differences are nonnormal and contain many extreme outliers, use the 1-sample Wilcoxon test (for symmetric data) or the 1-sample sign (for nonsymmetric data) to evaluate the paired differences.

Basic Statistical Tests

Do you have binary data or count data?

Binary data
Classifies items into one of two categories, such as pass/fail or yes/no. Often used to compare proportions.

Count data
Counts the presence of a characteristic, response, or activity. Often used to compare occurrence rates. Example An automotive manufacturer wants to compare the number of scratches on each car door that is produced by two production lines. Because the car doors have the same surface area, inspectors can easily compare the production lines by comparing the rate of scratches per door.

Focus of test: Compare groups with each other Number of groups: Two

Data type: Attribute

Example To evaluate a new delivery process, a delivery company counts the number of mishandled packages per day before and after it implements the new process. Because the total number of packages handled per day varies, proportions provide a better way to compare the number of mishandled packages.

Type of attribute data

Binary data

Count data

To collect count data, you count the number of occurrences in a given amount of time, area, volume, or other observation space. If a process has a constant rate of occurrence, use data from any observation space to make an inference about the entire process, such as the number of defects per day, per month, or per year. Otherwise, make sure that the observation space over which you collect the data is appropriate for the question you want to answer. Suppose the rate of phone calls at a call center varies greatly each hour. If you track only the calls received between 8 am and 9 am, the rate will not reflect the average hourly rate over the entire day. Conversely, if you track the calls over the entire day, the average hourly rate will not reflect the rate of calls received between 8 am and 9 am.

Basic Statistical Tests

Basic Statistical Tests

2 Proportions

2 Proportions
The 2 proportions test determines whether the proportions of two groups differ and provides a range of values that is likely to include the true difference. Example A delivery company tracks how many packages are mishandled each day before and after it implements a new delivery process. A quality analyst uses a 2 proportions test to determine whether the proportion of mishandled packages under the old process differs from the proportion of mishandled packages under the new process. To perform a 2 proportions test in Minitab, choose Stat > Basic Statistics > 2 Proportions.
Test and CI for Two Proportions

Focus of test: Compare groups with each other Number of groups: Two

Data type: Attribute

Sample 1 2

X 18 11

N 867 794

Sample p 0.020761 0.013854

Type of attribute data: Binary

Difference = p (1) - p (2) Estimate for difference: 0.00690734 95% CI for difference: (-0.00558971, 0.0194044) Test for difference = 0 (vs not = 0): Z = 1.08 P-Value = 0.279 Fishers exact test: P-Value = 0.349

2 Proportions

You can evaluate whether one proportion is greater than, less than, or not equal to the other proportion. If you choose a one-sided test (less than or greater than), the test has more power to detect a difference in the direction of interest. However, it cannot detect a difference in the opposite direction. Suppose you want to know whether a process change affects the proportion of mishandled packages. You could increase the power of the test by testing only whether the proportion for the new process is less than the proportion for the old process. However, this test will not detect a difference if the proportion for the new process is greater than the proportion for the old process. Use Power and Sample Size for 2 Proportions to determine how much data you need to detect an important difference between two proportions.

Basic Statistical Tests

Basic Statistical Tests

2-Sample Poisson Rate

2-Sample Poisson Rate

The 2-sample Poisson rate test determines whether the rates of two groups differ and provides a range of values that is likely to include the true difference. Example An inspector for an automotive manufacturer wants to compare the number of scratches per car door for two production lines. The inspector uses a 2-sample Poisson rate to evaluate whether the rate of scratches per door is the same for both production lines. To perform a 2-sample Poisson rate test in Minitab, choose Stat > Basic Statistics > 2-Sample Poisson Rate.
Test and CI for Two-Sample Poisson Rates: Line A, Line B

Focus of test: Compare groups with each other Number of groups: Two

Data type: Attribute

Variable Line A Line B

Total Occurrences 6 16

N 25 25

Rate of Occurrence 0.24 0.64

Type of attribute data: Count

Difference = rate(Line A) - rate(Line B) Estimate for difference: -0.4 95% CI for difference: (-0.767722, -0.0322782) Test for difference = 0 (vs not = 0): Z = -2.13 P-Value = 0.033 Exact Test: P-Value = 0.052 * NOTE * The normal approximation may be inaccurate for small total number of occurrences.

2-Sample Poisson Rate

You can evaluate whether one rate is greater than, less than, or not equal to the other rate. If you choose a one-sided test (less than or greater than), the test has more power to detect a difference in the direction of interest. However, it cannot detect a difference in the opposite direction. Suppose you want to detect whether a process change affects the rate of scratches per door. You could increase the power of the test by testing only whether the rate for the new process is lower than the rate for the old process. However, this test will not detect a difference if the rate for the new process is higher than the rate for the old process.

Basic Statistical Tests

Do you have continuous data or attribute data?

Continuous data
Measures a characteristic of a part or process, such as length, weight, or temperature. The data often includes fractional (or decimal) values.

Attribute data
Counts the presence of a characteristic or condition, such as a physical trait, a type of defect, or a rating, such as pass/fail. The data are whole numbers. Example An automated inspection process examines samples of bolts produced by four machines for severe cracks that make the bolts unusable. For each sample, analysts record the number of bolts that are inspected and the number of bolts that are rejected.

Focus of test: Compare groups with each other Number of groups: More than two

Example A quality engineer wants to evaluate whether the mean weight of cereal boxes is the same across three different production shifts. The analyst samples boxes from each shift and records their weights.

Data type

Continuous data

Attribute data

If possible, collect continuous data because they provide more detailed information. However, sometimes attribute data adequately describe the quality of a part or a process. For example, if you are tracking broken light bulbs, you dont need to measure a characteristic of the bulb to evaluate whether its broken or not. What matters is only the number of bulbs that are broken (counts).

Basic Statistical Tests

Do you want to compare the center or the spread of each distribution?

Center of distribution
Compares the central value of each group to determine whether at least one group differs from the others. Example A quality engineer wants to compare the hardness of four different paint blends after applying them to a metal surface. She wants to determine whether the hardness of at least one paint blend is different from the others.

Spread of distribution
Compares the variation of each group to determine whether the data vary more (or less) in at least one group than the others. Example A manager of a lumberyard measures the length of beams that are cut by three different sawmills. He wants to compare the spread of the measurements for each sawmill and determine whether the consistency of the beam lengths is the same.

Focus of test: Compare groups with each other Number of groups: More than two

Data type: Continuous

Focus of comparison

Center of the distribution

Spread of the distribution

You can evaluate the center of the data using the mean (the arithmetic average) or the median (the middle value when the data are arranged from lowest to highest). The median is less sensitive than the mean to outliers. You can evaluate the spread, or variation, of the data using the standard deviation or the variance. The standard deviation is often easier to interpret because it uses the same units as the data.

Basic Statistical Tests

Does your data follow a normal or nonnormal distribution?

Normal distribution
Data that follow a symmetric, bell-shaped distribution. Example A quality engineer measures the hardness of four different paint blends after applying them to a metal surface. The engineer determines that the hardness of each blend follows a normal distribution.

Nonnormal distribution
Data that follow a distribution that either is not symmetric or is symmetric but not bellshaped. Example A health administrator records the unoccupied bed space for three hospitals located in the same city. The administrator determines that the data are not symmetrically distributed and, therefore, are not normal.

Focus of test: Compare groups with each other Number of groups: More than two

Data type: Continuous

Focus of comparison: Center of the distribution

Data distribution

Normal distribution

Nonnormal distribution

The distribution of your data often depends on your process. For example, data that track cycle time for service processes, such as the time needed to process an application, serve a customer, or deliver a product, often do not follow a normal distribution. Although the normal distribution is always bell-shaped, not all bell-shaped distributions are normal. To determine whether your data follow a normal distribution, use normal probability plots, histograms, or normality tests such as the Anderson-Darling test. You can perform these evaluations using Minitabs Normality Test or Graphical Summary. If you have 30 or more observations in each sample, you can generally treat your data as being normal when you perform a basic statistical test to compare the means of two or more groups.

Basic Statistical Tests

Basic Statistical Tests

One-Way ANOVA

One-Way ANOVA
A one-way ANOVA determines whether the means of two or more independent groups differ and provides a range of values that is likely to include the true mean of each group. Example A quality engineer measures the hardness of four different paint blends after applying them to a metal surface. The engineer uses a one-way ANOVA to determine whether the mean hardness differs among the blends. To perform a one-way ANOVA in Minitab, choose Stat > ANOVA > One-Way. To determine which groups differ from one another, check Comparisons and select a method for comparing the groups.

Focus of test: Compare groups with each other Number of groups: More than two

Data type: Continuous

One-way ANOVA: Hardness versus Paint Source Paint Error Total DF 3 20 23 SS 281.7 312.1 593.8 MS 93.9 15.6 F 6.02 P 0.004

Focus of comparison: Center of the distribution

S = 3.950

R-Sq = 47.44% N 6 6 6 6 Mean 14.733 8.567 12.983 18.067 StDev 3.363 5.500 3.730 2.636

R-Sq(adj) = 39.56% Individual 95% CIs For Mean Based on Pooled StDev +---------+---------+---------+--------(-----*------) (------*------) (------*------) (------*------) +---------+---------+---------+--------5.0 10.0 15.0 20.0

Data distribution: Normal

One-Way ANOVA

Level Blend Blend Blend Blend

A B C D

Pooled StDev = 3.950

Typically, you use one-way ANOVA to evaluate the means of two or more groups if you have continuous data that are normally distributed. The test assumes that the data from each group have about the same spread (equal variance). To test the variances of the groups, use Minitabs Test for Equal Variances. You can also use one-way ANOVA to evaluate data that is nonnormal but meets one of these conditions: has 30 or more observations, is symmetric with at least 5 observations, or has relatively few extreme outliers.To quickly evaluate normality and outliers, use Minitabs Graphical Summary. If your data contains any extreme outliers, make sure they are valid measurements. Use Power and Sample Size for one-way ANOVA to determine how much data you need to detect an important difference between the group means.
Basic Statistical Tests

Basic Statistical Tests

Kruskal-Wallis

Kruskal-Wallis
The Kruskal-Wallis test determines whether the medians of two or more groups differ. Example A health administrator wants to compare the unoccupied bed space in three hospitals located in the same city. Because the data are not normal and contain extreme outliers, she uses the Kruskal-Wallis test to determine whether the median bed space in each hospital differs. To perform a Kruskal-Wallis test in Minitab, choose Stat > Nonparametrics > Kruskal-Wallis.
Kruskal-Wallis Test: Beds versus Hospital Kruskal-Wallis Test on Beds

Focus of test: Compare groups with each other Number of groups: More than two

Data type: Continuous

Focus of comparison: Center of the distribution

Hospital 1 2 3 Overall H = 7.05 H = 7.05

N 11 11 11 33

Median 16.00 31.00 17.00

Ave Rank 14.0 23.3 13.7 17.0

Z -1.28 2.65 -1.37

DF = 2 DF = 2

P = 0.029 P = 0.029

(adjusted for ties)

Data distribution: Nonnormal

Kruskal-Wallis

You can also use one-way ANOVA to evaluate data that is nonnormal but meets one of these conditions: has 30 or more observations, is symmetric with at least 5 observations, or has relatively few extreme outliers.To quickly evaluate normality and outliers, use Minitabs Graphical Summary. If your data contains any extreme outliers, make sure they are valid measurements. If you have many extreme outliers, you may want to use Moods median test to compare the central values of the groups. Although Moods median test is less powerful than the Kruskal-Wallis test, its results are not as strongly affected by outliers.

Basic Statistical Tests

Basic Statistical Tests

Test for Equal Variances

Test for Equal Variances

A test for equal variances determines whether the variation (variance or standard deviation) of two or more independent groups differs, and it provides a range of values that is likely to include the true variation of each group.

Focus of test: Compare groups with each other Number of groups: More than two

Example The manager of a lumberyard wants to compare the consistency of the length of beams cut by three different sawmills. He uses a test for equal variances to evaluate whether the variation of the beam lengths differs for any of the sawmills. To perform a test for equal variances in Minitab, choose Stat > ANOVA > Test for Equal Variances.

Data type: Continuous

Focus of comparison: Spread of the distribution

Test for Equal Variances

You can use the test for equal variances on normal or nonnormal data.

Basic Statistical Tests

Basic Statistical Tests

Cross Tabulation and Chi-Square

Cross Tabulation and Chi-Square

Cross tabulation and chi-square analysis tests whether the distribution of observations for an attribute variable differs for any of the groups. Example An automated inspection process examines samples of bolts from four suppliers and determines whether each bolt contains a severe crack that makes it unusable. Analysts use cross tabulation and chi-square to determine whether the distribution of cracked bolts differs among the suppliers. To perform cross tabulation and chi-square in Minitab, choose Stat > Tables > Cross Tabulation and Chi-Square. Click Chi-Square and check Chi-Square analysis.
Tabulated statistics: Supplier, Cracked bolt

Focus of test: Compare groups with each other Number of groups: More than two

Data type: Attribute

Rows: Supplier No Yes 4 8 10 0 22

Columns: Cracked bolt All 100 100 100 100 400 Count

Cross-Tabulation and Chi-Square

A B C D All

96 92 90 100 378

Cell Contents:

Pearson Chi-Square = 11.352, DF = 3, P-Value = 0.010 Likelihood Ratio Chi-Square = 16.026, DF = 3, P-Value = 0.001

For a more in-depth analysis of the relationship between the groups and their levels, consider using logistic regression.

Basic Statistical Tests

Contacts
Minitab World Headquarters
Minitab Inc.
Quality Plaza 1829 Pine Hall Road State College, PA 16801-3008 USA Training Phone: +1 814.238.3280 x3236 Fax: +1 814.238.4383 Email: [email protected] www.minitab.com/training Technical Support Phone: +1 814.231.2682 www.minitab.com/support; customer.minitab.com to log a question Mentoring Phone: +1 814.238.3280 x3236 Email: [email protected] www.minitab.com/training/mentoring/
Minitab is a global company with subsidiaries and representatives around the world. To find a Minitab partner in your country, visit www.minitab.com/contacts.

Training by Minitab maximizes your ability to improve quality. It helps you make more effective business decisions by teaching you how to analyze your data with Minitab Statistical Software and manage your projects using Quality Companion by Minitab.

Our specialists are highly skilled in Minitab software, statistics, quality improvement, and computer systems. Minitab subsidiaries and Independent Local Representatives around the world offer technical support by phone in their local language.

Mentoring by Minitab makes it easier to implement cost-saving quality improvement initiatives by providing the statistical support you need, just when you need it. We even begin with a free consultation.

2009 Minitab, Inc. All rights reserved. The contents of this publication may not be reproduced without permission. MINITAB and all other trademarks and logos for the Companys products and services are the exclusive property of Minitab, Inc. All other marks referenced remain the property of their respective owners. See www.minitab.com for more information.