24.12.2013 Views

Online Conjoint Analysis: - YouGov

Online Conjoint Analysis: - YouGov

Online Conjoint Analysis: - YouGov

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

<strong>Online</strong> <strong>Conjoint</strong> <strong>Analysis</strong>:<br />

The faster, the worse?<br />

Sabine Pützfeld (psychonomics AG)<br />

Dr. Torsten Melles (psychonomics AG)<br />

Bielefeld, March 22 nd 2006


Agenda<br />

I. Introduction: Characteristics of <strong>Conjoint</strong><br />

<strong>Analysis</strong><br />

II.<br />

Research objectives<br />

III. Methodology<br />

IV. Reliability during survey progress<br />

V. Effects of age and product category<br />

VI. Conclusions and directions for future research<br />

2


4 basic characteristics of <strong>Conjoint</strong> <strong>Analysis</strong> (I)<br />

1. Respondents evaluate multiattributive product<br />

stimuli.<br />

paired comparison of car 1 and car 2<br />

3


4 basic characteristics of <strong>Conjoint</strong> <strong>Analysis</strong> (II)<br />

2. Multiple decisions are<br />

made.<br />

4


4 basic characteristics of <strong>Conjoint</strong> <strong>Analysis</strong> (III)<br />

3. Product attributes are<br />

combined systematically.<br />

5


4 basic characteristics of <strong>Conjoint</strong> <strong>Analysis</strong> (IV)<br />

4. The utility of each attribute level („partworth“) is statistically<br />

derived from all judgments.<br />

Partworth 1<br />

+<br />

+<br />

Partworth 2<br />

Partworth 3<br />

<br />

Overall decision<br />

6


Agenda<br />

I. Introduction: Characteristics of <strong>Conjoint</strong><br />

<strong>Analysis</strong><br />

II.<br />

Research objectives<br />

III. Methodology<br />

IV. Reliability during survey progress<br />

V. Effects of age and product category<br />

VI. Conclusions and directions for future research<br />

7


Background: duration per paired comparison in seconds<br />

duration per paired<br />

comparison in seconds<br />

45<br />

40<br />

35<br />

30<br />

25<br />

20<br />

15<br />

10<br />

5<br />

0<br />

warm-up<br />

1 5 10 15 20 25<br />

survey progress (paired comparisons)<br />

On average, the first 5 to 6<br />

paired comparisons of a<br />

conjoint analysis take more<br />

time.<br />

<br />

Independent of the total<br />

number of paired<br />

comparisons, the duration<br />

of the following paired<br />

comparisons is nearly at the<br />

same constant level.<br />

study 1 (n=528)<br />

study 2 (n=163)<br />

study 3 (n=303)<br />

study 4 (n=306)<br />

study 5 (n=262)<br />

8


Problem Formulation<br />

Effect of exhaustion? – Reliability<br />

of the answers declines<br />

or<br />

Effect of routine? – Reliability of<br />

the answers does not decline<br />

9


Agenda<br />

I. Introduction: Characteristics of <strong>Conjoint</strong><br />

<strong>Analysis</strong><br />

II.<br />

Research objectives<br />

III. Methodology<br />

IV. Reliability during survey progress<br />

V. Effects of age and product category<br />

VI. Conclusions and directions for future research<br />

10


Methodology: Database<br />

Peer analysis of 5 <strong>Online</strong> <strong>Conjoint</strong> Analyses:<br />

(1) „checking account“ – November 2004 – 528 respondents from<br />

online panel – 27 paired comparisons (2 versions)<br />

(2) „wall paints“ – January 2005 – 163 respondents from online panel<br />

and recruitment by telephone – 18 paired comparisons<br />

(3) „mobile phone price plans“ – January 2005 – 303 respondents<br />

from online panel – 16 paired comparisons<br />

(4) „building society savings“ – February 2005 – 306 respondents<br />

from online panel – 20 paired comparisons<br />

(5) „car study“ – March 2005 – 262 respondents from online panel –<br />

27 paired comparisons<br />

Order of the paired comparisons was<br />

randomized in each study<br />

11


Methodology: <strong>Analysis</strong> of reliability<br />

Relying on the partworths (derived from responses to all paired comparisons) for<br />

each respondent, an individual pattern of preference will be generated.<br />

At a second stage, the partworths are used to calculate the theoretically correct<br />

answer of each paired comparison, provided that the statements are totally<br />

consistent.<br />

perfect consistency of answer = response on a paired comparison corresponds<br />

to the one calculated on the basis of the individual pattern of preference<br />

Measure of reliability: consistency of answer I<br />

correlation between response on a paired comparison corresponds to the one<br />

calculated on the basis of the individual pattern of preference<br />

Measure of reliability: consistency of answer II<br />

(squared) deviation of response on a paired comparison corresponds to the one<br />

calculated on the basis of the individual pattern of preference<br />

12


Agenda<br />

I. Introduction: Characteristics of <strong>Conjoint</strong><br />

<strong>Analysis</strong><br />

II.<br />

Research objectives<br />

III. Methodology<br />

IV. Reliability during survey progress<br />

V. Effects of age and product category<br />

VI. Conclusions and directions for future research<br />

13


Consistency of answer during survey progress I<br />

(correlation)<br />

consistency of answer<br />

(correlation between actual given answer and<br />

calculated answer)<br />

1<br />

0,9<br />

0,8<br />

0,7<br />

0,6<br />

warm-up effect<br />

<br />

<br />

<br />

Except for the first paired<br />

comparison the correlations<br />

of the considered studies<br />

are on a constantly high<br />

(0.84 – 0.92) level of<br />

consistency.<br />

No decline of consistency<br />

during survey progress<br />

Effect of exhaustion cannot<br />

be confirmed<br />

0,5<br />

1 5 10 15 20 25<br />

survey progress (paired comparisons)<br />

study 1 (n=528)<br />

study 2 (n=163)<br />

study 3 (n=303)<br />

study 4 (n=306)<br />

study 5 (n=262)<br />

14


<strong>Analysis</strong> of consistency of answers<br />

Relying on the partworths (derived from responses to all paired comparisons) for<br />

each respondent an individual pattern of preference will be generated.<br />

On a second step the partworths are used to calculate the theoretically correct<br />

answer of each paired comparison, provided that the statements are totally<br />

consistent.<br />

perfect consistency of answer = response on a paired comparison corresponds<br />

to the one calculated on the basis of the individual pattern of preference<br />

Measure of reliability: consistency of answer I<br />

correlation between response on a paired comparison corresponds to the one<br />

calculated on the basis of the individual pattern of preference<br />

Measure of reliability: consistency of answer II<br />

shows, how far the<br />

answer differs from<br />

predicted value<br />

(squared) deviation of response on a paired comparison corresponds to the one<br />

calculated on the basis of the individual pattern of preference<br />

15


Consistency of answer during survey progress II<br />

(deviation)<br />

first 6 paired comparisons of all considered studies<br />

6%<br />

5%<br />

rel. frequencies<br />

4%<br />

3%<br />

2%<br />

1%<br />

0%<br />

-3 0 3<br />

deviation of actual given answer and<br />

calculated answer<br />

16


Consistency of answer during survey progress II<br />

(deviation)<br />

last 6 paired comparisons of all considered studies<br />

6%<br />

5%<br />

rel. frequencies<br />

4%<br />

3%<br />

2%<br />

1%<br />

0%<br />

-3 0 3<br />

deviation of actual given answer and<br />

calculated answer<br />

17


Consistency of answer during survey progress II<br />

(deviation)<br />

first and last 6 paired comparisons of all<br />

considered studies<br />

6%<br />

<br />

No significant differences<br />

while comparing the<br />

distributions of the<br />

deviations regarding the<br />

first and the last 6 paired<br />

comparisons<br />

rel. frequencies<br />

5%<br />

4%<br />

3%<br />

2%<br />

<br />

<br />

Answers of both survey<br />

parts represent the same<br />

deviation from predicted<br />

value<br />

Effect of exhaustion cannot<br />

be confirmed<br />

1%<br />

0%<br />

-3 0 3<br />

deviation of actual given answer and<br />

calculated answer<br />

18


Agenda<br />

I. Introduction: Characteristics of <strong>Conjoint</strong><br />

<strong>Analysis</strong><br />

II.<br />

Research objectives<br />

III. Methodology<br />

IV. Reliability during survey progress<br />

V. Effects of age and product category<br />

VI. Conclusions and directions for future research<br />

19


Further questions<br />

Based on all respondents, no effects of the consistency<br />

of answer could be determined, which indicate a kind of<br />

exhaustion during progress of paired comparisons.<br />

Does the consistency of answer depend on the age of<br />

the respondent resp. do older people become<br />

exhausted earlier than younger people?<br />

Does the progress of consistency of answer depend on<br />

the product that is tested during the conjoint analysis?<br />

20


Age effects within consistency of answer<br />

1,6<br />

1,4<br />

older respondents<br />

answer less consistently<br />

on principle<br />

Averaged squared deviations of<br />

all paired comparisons<br />

older than 45 years<br />

1,17<br />

squared deviations<br />

1,2<br />

1<br />

0,8<br />

0,6<br />

0,4<br />

0,2<br />

0<br />

1 5 10 15 20 25<br />

survey progress (paired comparisons)<br />

up to 25 years (n=427)<br />

26 to 35 years (n=535)<br />

younger respondents<br />

answer more consistently<br />

on principle<br />

36 to 45 years (n=351)<br />

older than 45 years (n=244)<br />

36 to 45 years<br />

26 to 35 years<br />

up to 25 years<br />

<br />

21<br />

0,77<br />

1,05<br />

0,99<br />

An effect of exhaustion<br />

cannot be confirmed in any<br />

age group


Further questions<br />

Based on all respondents no effects of the consistency<br />

of answer could be determined, which indicate a kind of<br />

exhaustion during progress of paired comparisons.<br />

Does the consistency of answer depend on the age of<br />

the respondent resp. do older people become<br />

exhausted earlier than younger people?<br />

Does the progress of consistency of answer depend on<br />

the product that is tested during the conjoint analysis?<br />

22


Differentiation according to the tested product<br />

squared deviations<br />

2,5<br />

2<br />

1,5<br />

1<br />

0,5<br />

car study<br />

<br />

<br />

checking account<br />

Consistency of answer<br />

depends on the tested<br />

product (resp. on complexity).<br />

However no differences<br />

during the survey progress<br />

can be determined. That<br />

means no effects of<br />

exhaustion could be found in<br />

any of the researched<br />

studies.<br />

0<br />

1 5 10 15 20 25<br />

survey progress (paired comparisons)<br />

study 1 (n=528)<br />

study 2 (n=163)<br />

study 3 (n=303)<br />

study 4 (n=306)<br />

study 5 (n=262)<br />

23


Agenda<br />

I. Introduction: Characteristics of <strong>Conjoint</strong><br />

<strong>Analysis</strong><br />

II.<br />

Research objectives<br />

III. Methodology<br />

IV. Reliability during survey progress<br />

V. Effects of age and product category<br />

VI. Conclusions and directions for future research<br />

24


Conclusion<br />

The analyses indicate, that a decline in reliability during survey progress<br />

of an online conjoint analysis does not appear at all.<br />

Even after 27 paired comparisons, the answers of the respondents are<br />

as consistent as at the beginning of the survey. That applies to<br />

different age groups<br />

different tested products<br />

On the other hand a clear warm-up effect occurs, which confirms the<br />

results of former studies. The respondents require 2 to 3 tasks to reach<br />

the maximum of reliability within theirs statements.<br />

As long as no systematic deviation of the preference statement exists in<br />

this context (but merely a lower reliability), it is not necessary to exclude<br />

the first paired comparisons from the utility estimation. Consequently, it<br />

is not necessary to include special warm-up tasks.<br />

25


Directions for future research<br />

The reliability of the conjoint analysis results has been found to depend<br />

on the age of the respondents and on the tested product. That means:<br />

In order to draw conclusions with the same reliability, different amounts of<br />

data (case number, number of statements per respondent) according to the<br />

age group and to the product to be tested will be necessary. Therefore a<br />

further quantification would be reasonable.<br />

Considering age, one should investigate whether there is a possibility of<br />

creating the tasks for the older people in a different way, so that the answers<br />

will be more reliable.<br />

Considering product category, reasons for differences in reliability should be<br />

explored. Reasons may be e.g. the familiarity with the product or the<br />

involvement or the motivation to participate in the survey.<br />

26


Thank you for your attention!

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!