You are on page 1of 19

Case 3: Two-Way ANOVA

A Case Study
Presented to the Department of Industrial and Systems Engineering
De La Salle University
2nd Term, A.Y.. 2021-2022

In partial fulfillment
of the requirements of the course
Advanced Quantitative Methods Laboratory (LBYIE2C)

Case #3
Submitted by:
CARTAGO, Elaine Mari
ESPIRITU, Gwyneth Emerald
JACELA, Denise Joelle
MAPANOO, Darren Neil
PAPA, Alessandra Jasmine
Section EB2

Submitted to:
Sir Eric Sy
May 4, 2022
I. Objectives of the Experiment

This case consists of different objectives that will be the guide for the researchers
to accomplish the different requirements for this case. With the given data, the first
objective is to perform the pre-ANOVA diagnostics, ANOVA proper, and post ANOVA
analysis.

With the given first objective, the second objective is to perform the two-way
ANOVA which will help the researchers to gain data that will be used for the last
objective of the case.

The last objective of the case is to assess any significant differences in the
measurements of the wall thickness via the three different operators, the different part
items, and the interaction between the operator and the parts.

II. Experimental Plan (Factors, Levels, Procedure)

Response variable

The response, the dependent variable for this experiment, is the measured wall
thickness of individual parts with the use of a digital gauge.

Factors

The indicators for the experiment are the part and operator. The different
unidentified parts were measured (in inches) from the production lot. The three operators
for this experiment differ from one another in ways such as finger dexterity, learning
curves, and coordination of the person.

Levels

A total of ten levels will be managed. The ten individual parts will measure
randomly the wall thickness of an unidentified part in a production lot with the three
operators using the digital gauge.
Procedure

The tool, for the data collection for this case study, to be utilized is an identical
digital thickness measuring gauge. A simple data collection procedure will be conducted
for the three chosen operators. In a production lot, the operators are to record the wall
thickness (in inches) of 10 unidentified parts of the place in random order. A total of two
trials are to be made by each operator, for the measurement of the 10 random parts. The
measurements may be recorded by hand or through any device suitable for the operator.
The statistical tool two-way ANOVA will be used to observe and identify the difference
between the means of the groups.

III. Results and Analysis (Plots, ANOVA, and Diagnostics)

Figure 1. Descriptive statistics and individual plot for Operator 1


Based on Figure 1, the mean of the measured thickness by operator 1 is 0.9557.
The standard error means, standard deviation, and the individual plot in Figure 1 present
that operator 1 was consistent in the measurements of 5 parts, which are 2, 4, 5, 6, and 9;
but inconsistent in the rest with a deviation of one to two-thousandths of an inch.

Figure 2. Descriptive statistics and individual plot for Operator 2

Figure 2 and Figure 7 show that the mean of the measured thickness by operator 2
is higher than operator 1 which is equal to 0.95605. The individual value plot also
exhibits that the operator was not as consistent as operator 1 in measuring the thickness
of the wall considering only 4 of the measurements were unvarying, and the 6 remaining
have deviations.
Figure 3. Descriptive statistics and individual plot for Operator 3

Based on Figures 3 and 7, operator 3 recorded the highest mean for the
measurements of the wall thickness. Furthermore, it can be observed that operator 3 was
just as consistent as operator 2 with 4 uniform measurements, such as parts 5, 6, 8, and 9,
but the rest have deviations of one to two thousandths of an inch.
Figure 4. Probability plot: Thickness by Operator

Figure 4 presents the normality plot of measured thickness by the factor operator;
this shows that the recorded measurements (data points) by operator 1 are relatively close
to the fitted normal distribution line, while the data points by operators 2 and 3 are not
that close. In addition, the p-value of operator 1 is greater than the significance level of
0.05; on the other hand, the p-values of operators 2 and 3 are less than the significance
level. Therefore, the null hypothesis for operator 1 will be accepted and rejected for the
rest. This signifies that only the data points from operator 1 follow a normal distribution.
Figure 5. Probability plot: Thickness by Part

Same as Figure 4, Figure 5 also shows whether the data follow a normal
distribution or not. As it can be seen, about half of the measured thickness per part
follows a normal distribution since the data points are closer to the middle line.
Moreover, the p-values for parts 1, 2, 3, and 7 are greater than the significance level of
0.05; hence, the null hypothesis that states that data follow a normal distribution will be
accepted.

Figure 6. Bar chart of means: Thickness by Part


This bar chart compares the means per part measured by all of the operators. It
shows that the thickest part is part 6 with a measurement of 0.9577 inches and the least is
part 1 with 0.9538.

Figure 7. Bar chart of means: Thickness by Operator

Figure 7 compares the mean of the recorded thickness per operator and exhibits
that operator 1 recorded the lowest mean for the parts while operator 3 has the highest
mean with 0.95665.

Figure 8. Bar chart of means: Thickness by Part and Operator


Figure 8 compares the means of the recorded thickness measured by the
operators. Based on the chart, operators 1 and 2 have the same mean of measurements for
parts 3, 5, and 9. On the other hand, operators 1 and 3 recorded the same mean for parts 6
and 10. While operators 2 and 3 have the same means for parts 4 and 8. Tallying up the
times that the mean of measured thickness coincided with other operators: operator 1 and
operator 2 coincided with the other operators, 5 times, while operator 3 is in agreement
with the operators, 4 times. According to this, it can be concluded that operator 3 is not as
consistent nor reliable as the first two operators.

Figure 9. ANOVA Table with p-values

The analysis of variance showed that the p-value, for both part and operator, is
0.000 which is less than the significant level of 0.05. This means that the null hypothesis
which states that the part and operator have no linear relationship with the measured wall
thickness will be rejected, signifying that the part and operator have significant effects on
measured wall thickness. On the other hand, the interaction between part and operator has
a p-value of 0.490 which is greater than the significance level denoting that the null
hypothesis will be accepted, stating that the interaction between part and operator has no
significant effect on the measured wall thickness.
Figure 10. Box plot: Thickness by Operator

The box plot in Figure 10 shows that operator 3 mostly measured higher wall
thickness than the two other operators. Furthermore, all of the operators were only
consistent in their measurements about half of the time. Overall, the measurements for
parts 5, 6, and 9 were consistent or similar with all of the operators.

Figure 11. Box plot: Thickness by Part


Figure 11 is the same as Figure 10 except it shows the comparisons of the boxplot
by the variable–part–in a horizontal manner. This figure clearly shows the deviations or
inconsistencies made by the operators in the measurements per part. As it can be
observed, operator 1 recorded the largest deviation in measuring part 8, operator 2 in
measuring part 10, while operator 3 recorded the largest deviation in measuring part 1.

Figure 12. Tukey Pairwise Comparison of measurement of operators


Figure 13. Fisher Pairwise Comparison of measurement of operators

Aforementioned is the ANOVA test that was performed, the operator had a
significant relationship with the wall thickness thus, both Tukey Pairwise comparison and
Fisher pairwise comparison are executed in order to observe the mean differences
between the three operators. Observing the grouping information and the accompanying
graph by utilizing the Tukey pairwise comparison in fig. 12, the differences between the
three operators were shown to be not statistically significant due to the fact that all three
mean differences include zero in their confidence intervals. On the other hand,
employing the stricter parameters by the Fisher pairwise comparison, it is clear in fig. 13
that operators 1 and 3 have a significantly differing mean wall thickness from one
another.
Figure 14. Tukey Pairwise Comparison of measurement per parts
Figure 15. Fisher Pairwise Comparison of measurement per parts

In addition to the observations made on the comparison between the


measurements of the operators; measurements per part were also discerned by using the
Tukey and Fisher pairwise comparison once again. As seen in Figure 14, the Tukey
pairwise comparison identified four categories with parts that have similar mean
measurements. The four categories are also observed in fig. 15. In both methods, it is
seen that part 6 has the largest mean thickness while part 1 has the lowest.
Figure 16. Two-way ANOM for wall thickness

The fig. 16 further shows the interaction effect between the measurements of the
operator and part as well as, the main effects of the two, individually. Looking at the main
effects for parts; in comparison to the overall mean, parts 1, 9, and 10 exhibited a
significantly low thickness measurement while parts 4, 5, 6, and 7 show a significantly
high thickness measurement. Meanwhile, the main effects for operators show that only
operator 2 has a measurement value close to the mean, while operators 1 and 3 are shown
to have significantly low and high measurements, respectively. On the other hand,
interaction effects showed no statistical difference in the mean measurement of the wall
between the operator and part.

The ANOM for wall thickness verifies the results obtained in the Tukey and
Fisher pairwise comparison.
Figure 17. Interaction plot for wall thickness

Previously mentioned, the interaction effect between the operator and parts have
no significant effect on the measurements of the wall thickness. This is further proved by
the interaction plot in fig. 17 wherein, no evident pattern is observed.

Figure 18. Main effects plot for wall thickness


Meanwhile, the main effects plot shows that both operator and part have a
significant effect on the wall thickness, individually. It can be seen that operator 3 has the
highest effect on the wall thickness while operator 1 has the least. On the contrary, parts 6
and 1 have the highest and lowest effect on the wall thickness, respectively.

Based on the ANOVA table in fig.9, individual p-values were identified for both
operator and part additionally, the p-value for the interaction between the two was also
determined. Since α = 0.05, the lone p-values for operator and part, which are equivalent
to zero, are deemed significant to the wall thickness. However, the interaction between
the two has a p-value higher than 0.05 thus, it has no significant effect on the wall
thickness. In interpreting the significance results, the skill set of the operator in terms of
obtaining precise and accurate measurements should be considered as well as the
thickness of the parts that are being measured. Deeming the interaction between the parts
and operators significant would mean that even if the assessment in the measurement is
hasty and imprecise, the value obtained shall still be considered as significant to the wall
thickness.

Seeing that the interaction between the operator and parts measured is not
statistically significant, it can be concluded that the two operators will have no consistent
and reproducible results with one another. In addition to this, the three operators have
shown different main effects from one another wherein, operator 1 measured values
significantly lower than the meanwhile, operator 3 obtained measurements that are
significantly higher than the mean value. In terms of measuring the wall thickness,
operator 2 showed the nearest average measurement to the mean thus, operator 2 is the
best measurer of the three. Meanwhile, operator 1 has the highest deviation from the
mean hence, the least reliable among the three. Since both operator and parts are both
significant, the reliability of measurement can be observed by comparing the overall
mean to the average measurement obtained per part and the operators. The closer the
value obtained to the mean, the more reliable the measurement is.
IV. Conclusion

The objective of this experiment was to determine whether there were any
significant differences in the measurements of wall thickness for each operator, part item,
and the interaction between the operator and part using two-way ANOVA. Based on the
results of the experiment, both factors (part and operator) were found to be significant. As
their respective p-values were equal to zero (and less than 0.05), there is zero probability
that the average measurements via the operator or part items are equal, and although the
differences in measurements may seem small (as evident in their means and medians),
they are, nonetheless, significant. On the other hand, the part-operator interaction was
found to be insignificant. This is based on the graphs provided (two-way ANOM and
interaction plots) and the fact that the p-value (p = 0.490) for the interaction is greater
than α = 0.05. Furthermore, this lack of interaction between operator and part suggests
that the measurements for each part would still be the same (in this case, inconsistent), no
matter which operator is doing the inspecting. Lastly, changing the operator as well as
part items would affect the average wall thickness separately. Overall, since the
variabilities for each factor are quite high, the group concludes that this indicates a
negative effect on the reliability of the measuring system used.
V. List of References/Bibliography

Minitab 18 Statistical Software. What is a general linear model?. State College, PA:
Minitab, Inc. (www.minitab.com)

Minitab 18 Statistical Software. Understanding Analysis of Means. State College, PA:


Minitab, Inc. (www.minitab.com)

Minitab 19 Statistical Software. Example of an individual value plot of one Y variable


with groups. State College, PA: Minitab, Inc. (www.minitab.com)

Minitab 19 Statistical Software. Example of Boxplot. State College, PA: Minitab, Inc.
(www.minitab.com)

Minitab 20 Statistical Software (2020). [Computer software]. State College, PA: Minitab,
Inc. (www.minitab.com)

You might also like