Mobile devices such as tablets and smartphones are the survey platforms of the future. At SKIM, we’re preparing for this future by adapting survey interaction design to the specifics of mobile; shorter and more engaging choice exercises, with fewer choice tasks and fewer options per task. However, when doing this, we must make sure that the results of a mobile Choice-Based Conjoint (CBC) study are valid; as in, it delivers the same results as the same CBC study would on a traditional platform.
John and Gerard shared the results in empirical comparative studies between mobile and traditional CBC, to determine if and when the results of a mobile CBC are valid. Based on these results, they provided guidelines to determine when it is right to utilize mobile CBC.
For more information about SKIM's webinars, visit www.skimgroup.com/webinars.
How to Get Started in Social Media for Art League City
Webinar Mobile Conjoint Analysis: Are the results valid?
1. Share your thoughts online:
#SKIMwebinar
Mobile Choice-based Conjoint
A valid alternative
Gerard Loosschilder
CMethO
John Ashraf
Conjoint expert
Abigail Joffre
Today’s webinar host
2. Mobile surveys: it’s time to act
Increasingly people take surveys on a
mobile platform. Filtering them out
may skew the sample.
In
Standard
survey design
7%
4%
% of people taking
the test survey by
platform
Smartphone
89%
Tablet
PC
Mobile survey
design
Screen
Source: Burke, July 2013
Out
Biased sample
2
2013
SKIM Webinar “Mobile Choice-based Conjoint: A valid alternative”
3. But we should do a better job tailoring to their needs.
The respondent may have something better to do
Survey is
enjoyable
PC
24%
Tablet
Mobile ST
Mobile+
9%
PC
This is
bad news
Mobile ST
Mobile+
PC
79%
Tablet
14%
10%
Would take another
survey on the same
device
Responding
was very easy
53%
38%
31%
This is
bad news
Tablet
66%
48%
Mobile ST
35%
Mobile+
This is
bad news
32%
Source: Burke, July 2013. Mobile ST is a standard survey on a smartphone; Mobile+ is a dedicated mobile survey design
3
SKIM Webinar “Mobile Choice-based Conjoint: A valid alternative”
4. Today’s presentation is about the
validity of mobile choice-based conjoint
Introducing 3*3 mCBC
4
SKIM Webinar “Mobile Choice-based Conjoint: A valid alternative”
5. Adapting to the shorter attention span of the mobile user
By reducing the number of choice tasks and number of choices per task
Three choice tasks
5
Three options each
SKIM Webinar “Mobile Choice-based Conjoint: A valid alternative”
Completed in
6. At the same valid results as a traditional CBC exercise
The good news
The bad news
The MAE of share predictions is
<4% at the correct sample size
So far, it is impossible to reduce
the number of choice tasks and
options per task to fewer than the
3*3 platform
Overall results, conclusions and
recommendations are the same
6
SKIM Webinar “Mobile Choice-based Conjoint: A valid alternative”
7. 3*3 mCBC requires a larger sample size
As a rule of thumb, sample size is
tied to number of parameters to be
estimated, determined by the
number of attributes and levels
Traditionally, we recommend a minimal
sample size of n= 200 for any conjoint
study. Now, the sample size goes up
with the number of parameters to be
estimated.
7
If # of
parameters is
Sample size
n=
12 (3 atts*4 levels)
400
16 (4 atts*4 levels)
800
20 (5 atts*4 levels)
1,200
25 (5 atts*5 levels)
1,600
SKIM Webinar “Mobile Choice-based Conjoint: A valid alternative”
8. 3*3 mCBC requires a well-balanced research design
In a 3*3 mCBC research design,
the number of data points
collected is much smaller than the
number of parameters to be
estimated.
So, the key success factor is a
high-quality research design
8
Frequency balanced
Every combination of two
attributes and their levels is
represented well by all research
designs
Market knowledge included
We take current market shares
and knowledge about consumer
preferences into account in design
generation
SKIM Webinar “Mobile Choice-based Conjoint: A valid alternative”
9. Empirical evidence
How do we know if mCBC
using a 3*3 design is valid?
9
SKIM Webinar “Mobile Choice-based Conjoint: A valid alternative”
10. Because we have validated it …
We determined the relation between
sample size and convergent validity
It shows that at increasing sample sizes,
mobile CBC is a good approximation of
common CBC on traditional platforms
10
SKIM Webinar “Mobile Choice-based Conjoint: A valid alternative”
11. … in a study into pain killers and batteries
Attributes:
Brand
Admin method
Dosage
Package
Price
3-5 levels each;
16 parameters
Attributes:
Brand
Price
Size
3-5 levels each;
12 parameters
What is the validity of mobile CBC at various sample sizes
if compared with the same conjoint study in a traditional way at n= 200?
11
SKIM Webinar “Mobile Choice-based Conjoint: A valid alternative”
12. Attribute importance values show great similarities
at all sample sizes for mobile CBC
The values for
batteries are
equally similar
Attribute importance
Pain killers
40%
30%
Brand
20%
Admin method
10%
Dosage
0%
Package
Price
Variations in
attribute
importance values
are insignificant
Platform and sample size
12
SKIM Webinar “Mobile Choice-based Conjoint: A valid alternative”
13. However, at larger sample sizes, preference shares get
closer to benchmark leg, demonstrating greater validity
10.0%
Mean Absolute Error values (MAE, in %)
MAE value in %
7.5%
The break point is at n= 400; twice the
sample size of the benchmark
5.0%
Batteries
Pain killers
2.5%
Especially at pain killers at 16
parameters (instead of 12 for batteries)
0.0%
200
13
300 400 500 600 700 800
Sample size in the mobile leg
Mean Absolute Error
(MAE) is a metric
expressing how far
apart the preferences
shares are as
measured by mCBC at
increasing sample
sizes compared with
the benchmark CBC
at n= 200
900
SKIM Webinar “Mobile Choice-based Conjoint: A valid alternative”
14. Conditions to make 3*3 mCBC work – research design
Every display of product profiles in a choice task is valuable. It
should not be wasted on prohibitions or alternative specific designs.
• We can not tolerate any reduction in D-efficiency values
So for now, we can apply mCBC in some standard situations.
We do not apply 3*3 mCBC in situations in which there are
prohibitions, utility balanced designs and alternative-specific
designs
14
SKIM Webinar “Mobile Choice-based Conjoint: A valid alternative”
15. Conditions to make 3*3 mCBC work – real estate
There is no space for large
amounts of information
Max. 5 attributes
Text is succinct, keywords
only
No complex descriptions
Pictures are
icons or schematic
Be careful with package or
ad testing
15
SKIM Webinar “Mobile Choice-based Conjoint: A valid alternative”
17. 3*3 mCBC in Price Sensitivity Analyses – batteries
http://tinyurl.com/skimmcbc
17
SKIM Webinar “Mobile Choice-based Conjoint: A valid alternative”
18. 3*3 mCBC in Price Sensitivity Analyses – pain killers
http://tinyurl.com/skimmcbc
18
SKIM Webinar “Mobile Choice-based Conjoint: A valid alternative”
19. 3*3 mCBC in Telecom – design of a postpaid portfolio
http://tinyurl.com/skimmcbc
19
SKIM Webinar “Mobile Choice-based Conjoint: A valid alternative”
20. Mobile in MaxDiff applied to Semantics
and other tests of claims and benefit statements
http://tinyurl.com/skimmcbc
20
SKIM Webinar “Mobile Choice-based Conjoint: A valid alternative”
21. To conclude
3*3 mCBC is a valid addition
21
SKIM Webinar “Mobile Choice-based Conjoint: A valid alternative”
22. To conclude – mCBC is a valid addition
Do
Don’t
• Take market knowledge into
account
• Use it for large CBCs
• Use a larger sample
• Include research design
restrictions
• Use succinct text, iconized
visuals
• Use it for package, ad or
concept work
• Have five attributes at five
levels max
22
SKIM Webinar “Mobile Choice-based Conjoint: A valid alternative”
23. Next steps
• SKIM researchers will continue to validate mCBC …
• For allocation studies
• For large research designs using consideration sets
• For more complex visuals
• Keep improving UI for Mobile CBCs
• Across dominant platforms
• Translate learnings back to any CBC
23
SKIM Webinar “Mobile Choice-based Conjoint: A valid alternative”
24. Share your thoughts online:
#SKIMwebinar
Go to www.skimgroup.com/webinars
for today’s presentation slides and more!
Gerard Loosschilder
John Ashraf
Abigail Joffre
Chief Methodology Officer
g.loosschilder@skimgroup.com
@gloosschilder
Conjoint expert
j.ashraf@skimgroup.com
a.joffre@skimgroup.com
+1 201 963 8430
@SKIMgroup