This is a Factorial experimental design, aiming to answer two research questions:
(1) “Do people identify more with objects they own than with objects they access?”, and
(2) “Does thinking about owned objects make people act more altruistic?”.
The qualtrics analysis in this paper only applies for the parts designed to answer the first research question.
TataKelola dan KamSiber Kecerdasan Buatan v022.pdf
Experimental Design: Does people identify more with objects they own than with objects they access?
1. GROUP A1 |ASSIGMENT #2 1
IESEG Paris
MARKETING RESEARCH METHODOLOGY
OCTOBER 24, 2017
Professor: Bart CLAUS
Group Assignment #2
Experimental Design
GROUP A1:
LI Yan
ZHANG Siwei
NGUYEN Hoang Quan
NGUYEN Thi Duc Hanh
TRANG Thi Ngoc Thao (Traci)
Research Question 1:
Does people identify more with objects they own than with objects they access?
• Draw hypotheses
• Analyze the qualtrics design
• Analyze the data
• Draw critical conclusions
I. HYPOTHESIS
• DV: How people identify
(How much on the given scale that people identify with the car described.)
• IV1: Imagine of owning
(Context of car owning including 3 groups:
o Group 1: Renting
o Group 2: Owning
o Group 3: Uber)
• IV2: Gender
(including 2 groups:
o 1: Female
o 2: Male)
• Samples are independent: Respondents were divided into 3 different groups
‘Imagine of owning status’ to respond to identity questions related to their
assigned group
• H0
1: There is no difference in how people identify with objects they own and
objects they access
• H1: There is at least one difference in the way people identify with objects they
own and objects they access
• H02: There’s no interaction between gender and the way people identify with
objects they own and objects they access
• H2: There’s interaction between gender and the way people identify with objects
they own and objects they access
2. GROUP A1 |ASSIGMENT #2 2
II. QUALTRICS DESIGN ANALYSIS
This is a Factorial experimental design, aiming to answer two research questions: (1) “Do
people identify more with objects they own than with objects they access?”, and (2)
“Does thinking about owned objects make people act more altruistic?”. The qualtrics
analysis in this paper only applies for the parts designed to answer the first research
question.
The respondents to this experimental design, specified gender and age, were categorized
into 3 groups, depending on the context of their car owning. They were asked to describe
the car accordingly; then used the scale from 0 to 100 to measure how much they
identify with the previously-described car (scaling how relevant they find the statements
are to their identification.) The data, collected from 240 samples, met the rule of thumb
for 30 respondents per condition. Researcher can consequently use IV1: Imagine of
owning and IV2: Gender to test the influence on DV: How people identify.
First, for the survey purpose. Although the purpose of the researcher is clear and the
variables in the experimental design are target to answer the research, the authors find it
would have been better if the topic of the survey is communicated to the respondents
before it was spread out. Not knowing the purpose of the question, respondents
responded at length to describe the car, the data of which is not useful for researcher to
answer the research question.
Next, the data. The authors find the rich data collected (many samples) is helpful for a
more non-biased mean. At the same time, though the data is abundant, the design also
includes many other factors to the experimental design, tracking too many things at once:
Occupation, Age, Salary, Annual Income, creating different dimensions that make it more
complex to conclude which factor significantly affects the DV.
In addition, controlled variables: Although the variables are controlled: Age, Occupation,
Time doing the survey, Time spending on survey, the number of respondents for each
variable is not in good control (for example, Occupation, see Table1 below), nor there was
a perfect balance between test groups (ImagineOwing: Renting, Owning, Uber) (even
though it was good enough for the test) (see Table2 below). In other words, as
3. GROUP A1 |ASSIGMENT #2 3
respondents were selected randomly to take the survey (no target respondents defined),
and the decision on final data collected was unclear (decision to take responses from
240 samples out of 324), resulted in unequal number of respondents assigned to each
covariate (Age, etc.)
Table 1. (Occupation)
OCCUPATION
#
Respondents
%
Employed full time 50 21%
Employed part time 11 5%
Unemployed looking for work 3 1%
Unemployed not looking for work 2 1%
Retired 0 0%
Student 173 72%
Disabled 0 0%
Table2. (ImagineOwing)
Between-Subjects Factors N
GroupByImangine 1 89
2 71
3 80
Moreover, on the variables, the authors find there was an illogic point in the experimental
design with Age variable. The design defined the valid input value is from 16 to 62, which
is not logic, consider that it is illegal to rent or own a car below 18 years of age. In fact,
there was 6 respondents responded under 18 year-old.
Finally, scale that measured how people identify. The authors find there is too wide a gap
between two ends of the scale (from 0 to 100, with single unit is 1.) Without any
definition on each level, there is no guidance for respondents to evaluate the relevancy of
each statement (i.e. no label on each level of how much people identify with the car in
question), each respondent thus has different way to rate the scale. This affects the
quality of data.
As far as rating scales are concerned in an experimental design, there are several goals
that they should meet, in order to ensure the accuracy and effective of the data collected.
The scales should be defined in a way that is transparent for all respondents to
understand identically. It is necessary that the scale is reliable in the sense that if asked
again, the respondent should provide the same answer. The scale’s points should make it
easy to map out the meaning and for researcher to interpret. (Qualtrics , 2015)
That being said, the number of scale points depends on the type of question we are
asking. For the question this research was asking (How they identify), response is from
zero (not identify at all) to positive (absolutely identify), it is suggested to design a 5-point
scale, for example: Not at all – Slightly – Moderately – Very – Extremely. (Qualtrics , 2015)
4. GROUP A1 |ASSIGMENT #2 4
III.DATA ANALYSIS
Step 1: Data preparation
• Checked all measurement levels of variables
• Checked all value labels
• Sample size: 240 samples in total with more than 30 samples for each different condition.
Thus, size of sample is good enough according to Rule of Thumb
• Missing values check: no missing value
• Data coding:
Gender Imagine of Owning Respondent was
1 1 Woman who imagines to rent a car
1 2 Woman who imagines to own a car
1 3 Woman who imagines to use Uber
2 1 Man who imagines to rent a car
2 2 Man who imagines to own a car
2 3 Man who imagines to use Uber
Step 2: Run ANOVA test
Between-Subjects Factors
Value Label N
Gender 1 Female 151
2 Male 89
GroupByImangine 1.00 89
2.00 71
3.00 80
Levene's Test of Equality of Error Variances
a
Dependent Variable: IdentifyMean
F df1 df2 Sig.
2.246 5 234 .051
Tests the null hypothesis that the error variance of the dependent variable is equal across groups.
a. Design: Intercept + Gender + GroupByImangine + Gender * GroupByImangine
• From Levene’s Test, p= .051 > .05 which is not significant. So, we cannot reject
H0
1 yet.
5. GROUP A1 |ASSIGMENT #2 5
Tests of Between-Subjects Effects
Dependent Variable: IdentifyMean
Source
Type III Sum of
Squares df Mean Square F Sig.
Corrected Model 6620.910
a
5 1324.182 2.657 .023
Intercept 570589.675 1 570589.675 1144.903 .000
Gender 644.796 1 644.796 1.294 .257
GroupByImagine 5867.986 2 2933.993 5.887 .003
Gender * GroupByImagine 2015.657 2 1007.829 2.022 .135
Error 116619.499 234 498.374
Total 722737.181 240
Corrected Total 123240.408 239
a. R Squared = .054 (Adjusted R Squared = .034)
Gender
Dependent Variable: IdentifyMean
What is your gender? Mean Std. Error
95% Confidence Interval
Lower Bound Upper Bound
Female 49.280 1.819 45.696 52.864
Male 52.709 2.403 47.974 57.444
GroupByImagine
Dependent Variable: IdentifyMean
GroupByImangine Mean Std. Error
95% Confidence Interval
Lower Bound Upper Bound
1.00 46.509 2.422 41.737 51.281
2.00 58.540 2.831 52.963 64.117
3.00 47.934 2.562 42.887 52.981
Gender? * GroupByImagine
Dependent Variable: IdentifyMean
What is your gender? GroupByImangine Mean Std. Error
95% Confidence Interval
Lower Bound Upper Bound
Female 1.00 46.961 3.038 40.976 52.947
2.00 52.355 3.222 46.007 58.703
3.00 48.524 3.189 42.241 54.807
Male 1.00 46.057 3.773 38.623 53.492
2.00 64.725 4.655 55.554 73.896
3.00 47.344 4.010 39.445 55.244
6. GROUP A1 |ASSIGMENT #2 6
• From ANOVA test results, we have p value of variable – GroupByImagine p = .003
< .05. This indicates that there is significant effect of GroupByImagine on how
people identify.
• However, p value of variable – Gender p =.257 > .05, showing that effect of
gender on how people identify is not significant.
• On the other hand, p value of Gender*GroupByImagine p =.135 > .05, indicates
that these 2 IVs are statistically not interacted. So we can not reject H02.
From this step, we can conclude:
• The main effect is that: there is a difference in how people identify among
different groups divided by context of car owning (Imagine of owning.) For this
reason, we can reject H0
1.
• However, how different is not clearly indicated here. Therefore, it is necessary to
conduct Post-hoc test in order to find out the pattern of the difference.
• From our judgement, there might be an effect of Gender on how people identify
among different groups divided by context of car owning (Imagine of owning.) We,
therefore, drew a chart to see visually (see Chart 1 below).
• From the chart below, it is clear that there’s interaction between 2 IVs in the
second GroupByImagine (Owning) which is statistically shown as not significant
from previously-mentioned result.
• We, therefore, need to see Contrast Result to determine clearly.
Chart 1
0
10
20
30
40
50
60
70
ImagineRenting ImagineOwing ImagineUber
Males Female
7. GROUP A1 |ASSIGMENT #2 7
• Post-hoc test to determine the difference pattern: GroupByImangine
• Bonferroni Post-hoc comparisons indicate that there’s a significant difference
(p= .019) in how people identify objects they rent (M= 46.61, SD = 23.25) and
objects they own (M= 56.36, SD = 19.97). However, no significant difference is
found between the 2 other groups, in details:
• Group 1 and group 3: p value = 1.00 > .05. So there’s no significant difference
between groups Renting and using Uber
• Group 2 and group 3: p value = .71 > .05. So there’s no significant difference
between groups Owning and using Uber
Multiple Comparisons
Dependent Variable: IdentifyMean
(I)
GroupByIm
angine
(J)
GroupBy
Imangine
Mean
Differenc
e (I-J) Std. Error Sig.
95% Confidence Interval
Lower
Bound Upper Bound
Bonferroni 1.00 2.00 -9.7562
*
3.55233 .019 -18.3220 -1.1904
3.00 -1.4611 3.43939 1.000 -9.7545 6.8324
2.00 1.00 9.7562
*
3.55233 .019 1.1904 18.3220
3.00 8.2951 3.63992 .071 -.4819 17.0721
3.00 1.00 1.4611 3.43939 1.000 -6.8324 9.7545
2.00 -8.2951 3.63992 .071 -17.0721 .4819
Based on observed means.
The error term is Mean Square(Error) = 498.374.
*. The mean difference is significant at the .05 level.
Descriptive Statistics
Dependent Variable: IdentifyMean
What is your gender? GroupByImangine Mean Std. Deviation N
Female 1.00 46.9614 22.84857 54
2.00 52.3549 21.20947 48
3.00 48.5241 22.01388 49
Total 49.1830 22.03688 151
Male 1.00 46.0571 24.17376 35
2.00 64.7246 14.12629 23
3.00 47.3441 26.00502 31
Total 51.3296 23.86962 89
Total 1.00 46.6058 23.24646 89
2.00 56.3620 19.96852 71
3.00 48.0669 23.48595 80
Total 49.9790 22.70793 240
8. GROUP A1 |ASSIGMENT #2 8
• Finding significance in Interaction
Contrast Results (K Matrix)
a
Contrast
Dependent Variable
IdentifyMean
L1 Contrast Estimate -12.370
Hypothesized Value 0
Difference (Estimate - Hypothesized) -12.370
Std. Error 5.661
Sig. .030
95% Confidence Interval for Difference Lower Bound -23.524
Upper Bound -1.216
a. Based on the user-specified contrast coefficients (L') matrix: male2 versus female2
Test Results
Dependent Variable: IdentifyMean
Source
Sum of
Squares df Mean Square F Sig.
Contrast 2379.219 1 2379.219 4.774 .030
Error 116619.499 234 498.374
• From test result, we realize that p value of Contrast p = .03 < .05, which indicates
that there’s a significant difference in the way 2 genders identify with objects they
own and objects they access. Female identify less with objects they own (M=
52.35, SD= 21.21) than Male (M= 64.82, SD= 14.13) does with the contrast
estimate of -12.37.
Step 3: Run ANCOVA test
• There’s data about the real owning situation of respondents to a car. This question
was asked at the beginning of the test and independent to identity associated to
the imagined owning status. We judge that there might be an effect of this real
owning situation and would like to test it. Therefore, from the data given, we
created a new variable named ‘GroupByReality’ which is coded with values
explained as below table:
Value label Using Uber Lisence Rent No. Of Samples
1 Yes Yes Yes 67
2 Yes Yes No 62
3 Yes No Yes 7
4 No Yes Yes 10
5 Yes No No 78
6 No Yes No 11
7 No No Yes 1
8 No No No 4
9. GROUP A1 |ASSIGMENT #2 9
Assess the effects and interactions of IVs on the DV while statistically controlling 4
different covariates.
• :
Covariate 1: AGE
Tests of Between-Subjects Effects
Dependent Variable: IdentifyMean
Source
Type III Sum
of Squares df Mean Square F Sig.
Corrected Model 6630.513
a
6 1105.086 2.208 .043
Intercept 79571.116 1 79571.116 158.992 .000
Age 9.604 1 9.604 .019 .890
Gender 654.371 1 654.371 1.308 .254
GroupByImangine 5877.561 2 2938.781 5.872 .003
Gender *
GroupByImangine
2023.897 2 1011.949 2.022 .135
Error 116609.895 233 500.472
Total 722737.181 240
Corrected Total 123240.408 239
a. R Squared = .054 (Adjusted R Squared = .029)
• With the tested Covariate ‘Age’, the Adjusted R Squared = .029 is smaller than the
initial Adjusted R Squared (= .034) and p = .89 > .05. This indicates that Age can
be dropped from the final analysis.
Covariate 2: OCCUPATION
Tests of Between-Subjects Effects
Dependent Variable: IdentifyMean
Source
Type III Sum
of Squares df Mean Square F Sig.
Corrected Model 6303.825
a
6 1050.637 2.130 .051
Intercept 11564.635 1 11564.635 23.447 .000
Occupation 7.619 1 7.619 .015 .901
Gender 804.735 1 804.735 1.632 .203
GroupByImangine 5440.258 2 2720.129 5.515 .005
Gender *
GroupByImangine
1813.980 2 906.990 1.839 .161
Error 114428.229 232 493.225
Total 722737.181 239
Covariate
1.Age
2.Occupation
3.Income
4.GroupByReality
10. GROUP A1 |ASSIGMENT #2 10
Corrected Total 120732.054 238
a. R Squared = .052 (Adjusted R Squared = .028)
With the tested Covariate ‘Occupation’, the Adjusted R Squared = .028 is smaller than
the initial Adjusted R Squared (= .034) and p = .90 > .05. This indicates that Occupation
can be dropped from the final analysis.
Covariate 3: INCOME
Tests of Between-Subjects Effects
Dependent Variable: IdentifyMean
Source
Type III Sum
of Squares df Mean Square F Sig.
Corrected Model 6859.950
a
6 1143.325 2.291 .037
Intercept 34913.001 1 34913.001 69.968 .000
Income 1400.789 1 1400.789 2.807 .095
Gender 1158.444 1 1158.444 2.322 .129
GroupByImangine 4929.067 2 2464.534 4.939 .008
Gender *
GroupByImangine
1204.046 2 602.023 1.207 .301
Error 103788.221 208 498.982
Total 677381.875 215
Corrected Total 110648.171 214
a. R Squared = .062 (Adjusted R Squared = .035)
With the tested Covariate ‘Income’, the Adjusted R Squared = .062 is bigger than the
initial Adjusted R Squared (= .034) while p = .95 > .05. This indicates that Income has
effect on the treatment effects and can be considered in the final analysis although its
effect is not significant. And there might be another covariate that improve R Squared
significantly but is not included in this test.
Covariate 4: GROUP BY REALITY
Tests of Between-Subjects Effects
Dependent Variable: IdentifyMean
Source
Type III Sum
of Squares df Mean Square F Sig.
Corrected Model 6632.056
a
6 1105.343 2.209 .043
Intercept 169729.909 1 169729.909 339.144 .000
GroupByReality 11.146 1 11.146 .022 .881
Gender 631.531 1 631.531 1.262 .262
11. GROUP A1 |ASSIGMENT #2 11
GroupByImangine 5865.229 2 2932.614 5.860 .003
Gender *
GroupByImangine
2026.223 2 1013.111 2.024 .134
Error 116608.353 233 500.465
Total 722737.181 240
Corrected Total 123240.408 239
a. R Squared = .054 (Adjusted R Squared = .029)
• With the tested Covariate ‘Group by Reality’, the Adjusted R Squared = .029 is
smaller than the initial Adjusted R Squared (= .034) and p = .881 > .05. This
indicates that ‘Group by Reality’ can be dropped from the final analysis.
IV.SUMMARY
In short, from the data analysis above for this experimental design, we drew out:
• How people identify objects they rent and objects they own are significantly
different. However, no significant difference is found between the 2 other groups
under research.
• With object people own, 2 genders identify significantly differently: Female identify
less with objects they own than Male does.
• Income has effect on the treatment effects although its effect is not significant.
And there might be another covariate that improve R Squared significantly but is
not included in this test.
Regarding qualtrics design, we detected some issues that could lead to ineffectiveness in
data analysis:
• The topic of the survey should have been communicated to the respondents
before it was spread out
• Data: too many things were tracked at once: Occupation, Age, Salary, Annual
Income, creating different dimensions that make it more complex to conclude
which factor significantly affects the DV
• The fact that respondents were selected randomly to take the survey and the
decision on final data collected was unclear resulted in unequal number of
respondents assigned to each covariate (Age, etc.)
• There was an illogic point in the experimental design with Age variable
• The gap between two ends of the scale that measured how people identify (from 0
to 100, with single unit is 1) was too wide without any definition on each level.
REFERENCES
Qualtrics . (2015, APR 22). Three Tips for Effectively Designing Rating Scales. Retrieved
OCT 31, 2017, from Qualtrics : https://www.qualtrics.com/blog/three-tips-for-effectively-
using-scale-point-questions/