Case 5: Fractional Factorial Designs

A Case Study
Presented to the Department of Industrial and Systems Engineering
De La Salle University-Manila
1st Term, A.Y. 2023-2024

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

Carosus, Christian Jeric D.

Esguerra, Julio Fernando G.
Gambong, Roddick Eumer V.
Yabut, Brian Christian G.
Zaragoza, Leigh Dominique C.
Section ER2

Sir Eric Siy

November 21, 2023
 I. Objectives of the Experiment

The primary objective of this case study is to investigate the effectiveness of a half-fraction
factorial experiment design in relation to the previously conducted full factorial experiment when
examining the performance of the Sigmazone Catapult. Using the same methods in Case 4 in generating a
randomized set of trials is intended to illustrate the fundamental principles of the fractional factorial
design. It enables the researchers to obtain statistically significant conclusions with a substantially
reduced number of experimental runs. This study aims to assess the reliability of the fractional factorial
approach in generating conclusions about the catapult’s performance by conducting a side-by-side
comparison with Case 4 which used a full factorial design.

II. Experimental Plan (Factors, Levels, Procedure)

3
Similar to the experimental plan in Case 4, a ½ Fractional Factorial Design with 2 blocks and 5
replicates were implemented, obtaining and facilitating a total of 40 trials instead of 80 trials. The low and
upper limits of each factor are taken into account and remain the same; the firing angle has a lower limit
of 100 and an upper limit of 125 that was set by the group. There are two upper and lower limits to the cup
elevation: 200 and 300, respectively. There are two upper and lower limits to the pin elevation: 100 and
200. There are two upper and lower limits to the bungee position: 100 and 200. 180 degrees will be the
constant value of the release angle. The experiment is also conducted using the same Single Cup Virtual
Catapult available on SigmaZone. To ensure that every member in the group completes 8 trials, the
experimental runs will be randomly divided into five blocks using the _Stat_DOE_Factorial function of
the Minitab software (changing the “Full Factorial” setting to “½ Fractional Factorial”). Boxplots for
significant components, half-normal plots of effects, Pareto charts of effects, individual data plots,
normality plots, ANOVA tables with p-values, and main effects plots are all examples of detailed
statistical analysis. finding trends, connections, and significant impacts in the data to provide a solid
foundation on which to draw insightful conclusions. These analyses aim to determine why and how
switching between a ½ Fractional Factorial Design and a Full Factorial Design matters by comparing this
set of data and graphs to Case 4.
III. Results and Analysis (Plots, ANOVA, Diagnostics)
1. Table 1: 40-trial data for catapult
a. Minitab data range settings

Release Angle Fixed at 180°

Firing Angle Lower Limit: 100°

Upper Limit: 125°

Cup Elevation Lower Limit: 200

Upper Limit: 300

Pin Elevation Lower Limit: 100

Upper Limit: 200

Bungee Elevation Lower Limit: 100

Upper Limit: 200

b. Block 1 (Trial #1 - #8)

Firing Cup Pin Bungee Distance

StdOrder RunOrder CenterPt Blocks Angle Elevation Elevation Position (cm)
2 17 1 1 125 200 100 200 181
6 18 1 1 125 200 200 100 205
7 19 1 1 100 300 200 100 290
8 20 1 1 125 300 200 200 621
5 21 1 1 100 200 200 200 265
4 22 1 1 125 300 100 100 182
1 23 1 1 100 200 100 100 106
3 24 1 1 100 300 100 200 284

c. Block 2 (Trial #9 - #15)

Firing Cup Pin Bungee Distance

StdOrder RunOrder CenterPt Blocks Angle Elevation Elevation Position (cm)
13 9 1 2 100 200 200 200 274
11 10 1 2 100 300 100 200 287
10 11 1 2 125 200 100 200 178
12 12 1 2 125 300 100 100 185
 9 13 1 2 100 200 100 100 109
16 14 1 2 125 300 200 200 618
15 15 1 2 100 300 200 100 293
14 16 1 2 125 200 200 100 202

d. Block 3 (Trial #16 - #24)

Firing Cup Pin Bungee Distance

StdOrder RunOrder CenterPt Blocks Angle Elevation Elevation Position (cm)
17 33 1 3 100 200 100 100 108
24 34 1 3 125 300 200 200 619
19 35 1 3 100 300 100 200 282
22 36 1 3 125 200 200 100 205
21 37 1 3 100 200 200 200 257
18 38 1 3 125 200 100 200 184
23 39 1 3 100 300 200 100 298
20 40 1 3 125 300 100 100 185

e. Block 4 (Trial #25 - 31)

Firing Cup Pin Bungee Distance

StdOrder RunOrder CenterPt Blocks Angle Elevation Elevation Position (cm)
30 1 1 4 125 200 200 100 200
25 2 1 4 100 200 100 100 108
27 3 1 4 100 300 100 200 284
31 4 1 4 100 300 200 100 290
26 5 1 4 125 200 100 200 181
29 6 1 4 100 200 200 200 262
32 7 1 4 125 300 200 200 626
28 8 1 4 125 300 100 100 186

f. Block 5 (Trial #32 - 40)

Firing Cup Pin Bungee Distance

StdOrder RunOrder CenterPt Blocks Angle Elevation Elevation Position (cm)
37 25 1 5 100 200 200 200 269
 36 26 1 5 125 300 100 100 186
34 27 1 5 125 200 100 200 177
33 28 1 5 100 200 100 100 109
35 29 1 5 100 300 100 200 287
40 30 1 5 125 300 200 200 606
38 31 1 5 125 200 200 100 202
39 32 1 5 100 300 200 100 302

2. Analysis of Variances (ANOVA)

a. Case 4 [Full Factorial Design] ANOVA Table:

b. Case 5 [½ Fractional Factorial Design] ANOVA Table and Alias Structure:

 The ANOVA table for Case 4/Full Factorial Design showcased 3 interactions (2-Way, 3-Way, and
4-Way) with 1 factor (Firing Angle) not being statistically significant. On the contrary, the ANOVA table
for Case 5 shows only 1 interaction (2-Way) along with all of its 4 factors being “significant” since their
p-value equates to “0.000” which is smaller than 0.05 (the indicator to identify whether a factor is
statistically significant or not). The reason the ½ Fractional Factorial Design setting for Case 5 effects in
showcasing the number of interactions may be because of the aliasing/confounding structure, a way of
describing a design's confounding pattern. Because of this, the ½ Fractional Factorial design does not
incorporate every possible combination of factor levels, fractional factorial designs can lead to aliasing,
sometimes referred to as confounding. The estimated effect for A is the total of the effects of A and BCD,
for instance, if factor A is aliased with the three-way interaction BCD. It is impossible to distinguish
between a significant effect that results from BCD, A, or from a mixture of the two. In addition to this, the
2-Way interactions for Case 5 only showcases if the “Firing Angle” factor is involved.

3. Comparing the DOE/Factorial Design Graphs between Case 4 and Case 5 data
a. Residual Plots

Case 4 [Full Factorial Design] Case 5 [½ Fractional Factorial Design]

The residual plots for both Case 4 and 5 data are quite similar, resulting in a normal distribution
in both normal probability plot and histogram along with outliers. The versus fits plot cluster groups for
Case 5 is shown to be more organized and aligned than Case 4’s data which is scattered. As for the versus
fits plot, they both showcase a somewhat similar pattern: clusters of cyclical-trend patterns and a
possibility of being deemed as irrelevant/nonconstant variance.

b. Pareto Chart of the Standardized Effects

Case 4 [Full Factorial Design] Case 5 [½ Fractional Factorial Design]

In the Pareto chart above, it is evident that all factors are significant. In contrast to the Full
Factorial Experiment, the Pareto chart for the Fractional Factorial Experiment indicates that all
individual factors and some combinations are deemed significant. This alone suggests that, within the
chart , each factor and certain interactions have an effect on the performance of the catapult. In this case,
Factor B remains the strongest factor, but Factors C and D have switched places in terms of significance.
In this case, Factor C is now the second most significant factor, followed by Factor D. The shift in the
significance in Factors C and D might indicate that these factors have different levels of impact on the
catapult’s performance in the context of the Fractional Factorial Experiment.

c. Half Normal Plot

Case 4 [Full Factorial Design] Case 5 [½ Fractional Factorial Design]

 Similar to the Pareto chart, the Half-normal plot for the ½ Fractional Factorial Experiment
indicates that all individual factors and certain combinations are considered significant. This suggests
that, according to the Half-normal plot, each factor and other interactions have a significant effect on the
performance of the catapult. The observation that all factors are now considered significant in the
Fractional Factorial Experiment, while in Case 4 B, C, and D were significant, highlights the substantial
impact of the experimental design. The Fractional Factorial Experiment, with less experimental runs,
effectively captures the significance of all factors, presenting a more efficient way to assess the primary
influences on the catapult’s performance. The ability to identify all factors as significant in the FRactional
Factorial Design indicates the success of the design in efficiently identifying essential effects with a
reduced number of experimental runs when compared to the Full Factorial Design.

3. ANOVA / Other graphs

a. Main Effects Plot

Case 4 [Full Factorial Design] Case 5 [½ Fractional Factorial Design]

 It is evident that there are differences in means in the Main Effects Plot for Distance compared to
the previous Case, indicating the changes in the performance of the Sigmazone Catapult. Specifically, all
factors have increased, especially the Pin Elevation. The change in means in Case 5 for all factors suggests
an overall improvement in average distance traveled by the virtual catapult. The particular emphasis on
the increase in means for Pin Elevation suggests that, in Case 5, changes in the elevation of the pin have
significant impact on improving the performance of the catapult. This may also indicate that optimizing
the Pin Elevation is crucial for achieving greater distances.

b. Interaction Plot

Case 4 [Full Factorial Design] Case 5 [½ Fractional Factorial Design]

 c. Box Plot

Case 4 [Full Factorial Design] Case 5 [½ Fractional Factorial Design]

 Comparing the boxplots of each factor, Case 5 does not have any outliers unlike Case 4 wherein
outliers are present in every factor except “Pin Elevation”. It is also noticeable that the upper limit for each
factor in Case 5 has a bigger range than Case 4 whilst the lower limit has a smaller range. The medians are
almost quite similar with the exception of the “Firing Angle” factor.

IV. Conclusion

The analysis of variance(ANOVA) findings reveal that the Full Factorial Design from case 4 and
the Half-Fractional Factorial Design from this case have found distinct sets of significant factors. In the
fourth case, three types of interactions (2-way, 3-way, and 4-way) were examined, and it was found that
one element which is the Firing Angle did not exhibit statistical significance. On the other hand, Case 5
exhibited a sole significant interaction (2-way) and established the statistical significance of all four
components. The observed discrepancies in these factors can be explained by the presence of aliasing or
confounding structure that is inherent in fractional factorial designs. The ½ Fractional Factorial Design
may not encompass all possible combinations of factor levels making it challenging to distinguish
between significant effects.

In terms of interactions, Case 4 demonstrated a greater number of interactions in comparison to

Case 5 which present a notable 2-way interaction. The observed distinction arises as a direct outcome of
the confounding characteristics inherent in fractional factorial designs, wherein certain interactions are
merged, rendering the isolation of individual effects challenging. The inherent tradeoff associated with
employing a fractional factorial design becomes apparent through the decreased number of experimental
runs, which enables enhanced efficiency but compromises the capacity to distinctly detect specific
interactions. The presence of aliasing caused by the fractional design poses a difficulty in accurately
identifying the specific origin of significance, as evidenced by the results of the analysis of variance.

In summary, the Half-Fractional Factorial Design (Case 5) exhibited efficacy in finding relevant
factors while minimizing the number of experimental runs required. Nevertheless, this enhanced
efficiency is accompanied by the drawback of diminished capacity to accurately discern specific
individual interactions. The tradeoff is dependent upon the researcher's priorities, with the fractional
design presenting a feasible alternative if prioritizing a more economical utilization of resources. The
fractional factorial designs have the advantage of being extremely efficient designs since they estimate
the effects using a subset or fraction of the entire set of possible design runs. The drawback is that not all
impacts can be calculated because they confound some effects during the process. However, in the case
when a thorough comprehension of individual interactions is of utmost importance, the Full Factorial
Design may be favored, even though it necessitates a greater number of experimental runs. When making
a decision between the two designs, it is important to carefully analyze the trade-off between limited
resources and the level of detail necessary to achieve the specific study objectives.

V. List of References/Bibliography

What are confounding and alias structure? - Minitab. (n.d.). Support. Retrieved November 21,

2023, from

https://support.minitab.com/en-us/minitab/21/help-and-how-to/statistical-modeling/doe/su

pporting-topics/basics/what-are-confounding-and-alias-structure/
Analysis of Variance table for One-Way ANOVA - Minitab. (n.d.). Support. Retrieved October 7, 2023, from

https://support.minitab.com/en-us/minitab/20/help-and-how-to/statistical-modeling/anova/ho

w-to/one-way-anova/interpret-the-results/all-statistics-and-graphs/analysis-of-variance-table/

Chapter 11 Two-Way ANOVA. (n.d.). Statistics & Data Science. Retrieved October 20, 2023, from

https://www.stat.cmu.edu/~hseltman/309/Book/chapter11.pdf

guide, s. (n.d.). The Ultimate Guide to ANOVA - Graphpad. GraphPad. Retrieved October 20, 2023, from

https://www.graphpad.com/guides/the-ultimate-guide-to-anova

Interpret the key results for Interaction Plot - Minitab. (n.d.). Support. Retrieved October 20, 2023, from

https://support.minitab.com/en-us/minitab/21/help-and-how-to/statistical-modeling/anova/ho

w-to/interaction-plot/interpret-the-results/key-results/

Residual plots for Fit General Linear Model - Minitab. (n.d.). Support. Retrieved November 14, 2023, from

https://support.minitab.com/en-us/minitab/21/help-and-how-to/statistical-modeling/anova/ho

w-to/fit-general-linear-model/interpret-the-results/all-statistics-and-graphs/residual-plots/

Two-way ANOVA in SPSS Statistics - Step-by-step procedure including testing of assumptions. (n.d.). Laerd

Statistics. Retrieved October 20, 2023, from

https://statistics.laerd.com/spss-tutorials/two-way-anova-using-spss-statistics.php