Identifying Market Opportunities - Market Research
Identifying Market Opportunities - Market Research
Identifying Market Opportunities - Market Research
Consumer products
Success ful Design(D) Average 0.5 * Probability Success ful Market Tes t Market(T) Succ ess(M) given D given T 0.45 * 0.85 Overall Succ ess = 0.19
The results suffer from survivor bias which inflates success rates. Hence, the success rate may be much much lower (e.g. G.E. 2%)
Erics team was going to do Videocart right using the latest technology. They conducted a careful concept test for the new video cart.
Retailers Reluctant to try another version of a product that had a terrible reputation. The team would have had to establish extensive beta sites. However, retailers did not want to pay for it. Manufacturers Wanted to know whether ads on the Videocart would increase sales before committing significant advertising and promotion funds to the new medium. This required extensive beta sites again. However, manufacturers did not want to pay for it.
Pseudo-Experiment I
CASE I A newly designed catalog of electronic texts was mailed to potential customers and the firm sold 20 percent more compared to the previous year.
Question Can we conclude that the new catalog increased sales?
Pseudo-Experiment I (contd)
No. We cannot rule out other factors (e.g. improved economy) What is the problem? No control group The mailing catalog was mailed to all potential buyers. How to remedy the problem? Divide potential buyers or cities randomly into two groups: Treatment Group Control Group Mail the new catalog Mail the previous catalog
Treatment
Pseudo-Experiment II
CASE II A firm has two independent subsidiaries, one for the East coast market (A) and the other for the West cost market (B). Each subsidiary makes its regional advertising expenditure decision independently. When A increased its advertising expenditure and Bs advertising expenditure remained the same, As sales increased 10% while Bs sales did not change.
Question Can we conclude that increased advertising expenditure increased sales?
Other factors affect both groups: we can infer the causal relationship between catalog design and sales by comparing sales between groups.
Pseudo-Experiment II
No. It is not clear which of the following is true:
Advertising expenditure causes sales to change. The change in sales caused advertising expenditure to change.
In Experiment,
We want to make a causal inference between treatment and outcome. In order to make valid causal inference,
1. There must be a control group that did not receive the treatment or received a different treatment. - To rule-out out competing explanation - To cancel-out uncontrollable events external to the experiment 2. There must be random assignment of treatments to groups. 3. The experimental treatment must precede the outcomes.
It is not clear whether the change in advertising activity preceded the change in sales.
Advertising Expenditure
Sales
Exercise 1
A young PC game company developed a new strategy simulation game and is considering inserting a rebate coupon into the package.
Exercise 1 (contd)
Randomized Control Group Design
Treatment Group (G1) Control Group (G2) Package with coupon Purchase intention (X) (O1) Package without coupon Purchase intention (O2)
How to test whether the rebate coupon is a good idea? Think about: Randomized groups(G)
Treatment (X; the rebate coupon)
Output (O; measurement of interest)
Then, design an experiment.
Exercise 2
The PC game company is considering three different product packages, A, B, and C. The company wants to pick the best among them. Design an experiment to help them. Note that the level of treatments is 3 (A,B,and C)
Group 1 (G1)
Group 2 (G2) Group 3 (G3)
Exercise 2 (contd)
Randomized Control Group Design
Package A (X1)
Package B (X2) Package C (X3)
The output of interest: max(O1,O2, O3) Here there is no separate control group.
Each group acts as a control group for the others.
Factorial Design
An ISP firm wants to test three marketing variables: price, message, and promotion.
1. Price (4 levels) $20/month $30/month $40/month $50/month 2. Message (2 levels) Fast download speed Fast customer service 3. Promotion (2 levels) free for the first month free for the external modem
Sample Size?
The size of each group does not need to be big. In the randomized control group design, 30 people in each group may be sufficient. In the fractional factorial design, 10~15 people in each group is sufficient. Remember that the most important thing is an appropriate control or balance to minimize noise.
G2
G3
Free modem Fast download speed X X
G4
Free modem Fast customer service X
First First month free month free Fast download speed X X X X Fast customer service
Limitation of Experiments
Laboratory Experiment vs Field Experiment Between-subject Design vs Within-subject Design Internal Validity and External Validity
Low-Cost Sampling
Focus groups Mall intercepts Quota sampling Convenience sampling Judgment sampling Snowball and network sampling How many respondent do we need? Several thousand? No 200 at most
If there is something wrong in the questionnaire, the survey data is just a waste
Use a focus group interview to understand important variables (e.g. product attributes) before writing about questions. Conduct a pre-test of the questionnaire (N=30)
Actionability
Measurement Scale
Ordinal
Rank order among brands in terms of preferences
Minimize the number of open-ended questions. Choice from a list of multiple responses is better than open-ended questions. Use simple, direct and familiar vocabulary to all respondents Avoid vague or ambiguous meanings
e.g. How long have you lived here?
Categorical
Yes/No choice Single or multiple choice(s) from a list of responses
Interval
Likert scale
Start with easy questions and put lengthy and difficult questions at the end. General questions first and specific questions later. Sensitive questions dealing with income status, skill level, and so forth should not be placed at the beginning.
Advantages
Fast and very cheap Accessible to diverse and large group of people Error free in terms of coding On-line statistical analysis Vivid stimuli (e.g. photo, graphics) can be used.
Disadvantages
No control on sampling
Statistical Analysis
If responses are categorical, use a cross-tab (x2 test) If responses are interval scaled,
Test differences between two groups: 2-sample t-test Test differences between 2 interval responses: 1-sample t-test Test differences among >2 groups: one-way ANOVA Test differences with moderator: ANCOVA Test multiple responses among groups: MANOVA, MANCOVA Others Correlation; regression; discriminant analysis; cluster analysis; logistic regression.