Direct Marketing Research and Experimentation

1
Direct Marketing Researchand Experimentation

• Chapter 11

2
Research Serves Direct Marketers

• Fact-Finding
• Information Gathering
• Problem-Solving
• Decision-Making

3
Need for Marketing Research

• Managers must have good, accurate, timely
  information with which to make decisions.
• Marketing research helps to gather the needed
  information.
• The results of research can be quantitative
  and/or qualitative.
• Valid research measures results not opinions.

4
Problem Structure

• How much advertising is needed?
• How will the direct marketing mix be selected?
• How will our resources be utilized?

5
The Nature of Research

• Surveys vs. Experiments
• A survey looks at things the way they are. A
  mailed questionnaire, for example, may attempt to
  profile respondents to a product offer or
  promotion strategy. It may seek to anticipate
  future buying intentions to determine product or
  service preferences to guide pricing decisions
  or to measure attitudes, interests, opinions.
• An experiment is designed to measure the effect
  of change. What is the effect of a product
  change? What happens when a price level is raised
  or lowered? What is the result of selective
  promotion to specific market segments? What is
  the response influence of one promotion strategy
  relative to others?

6
Databased Research and Analysis

• My minds made up. Dont confuse me with
  facts.
• A characteristic of database-driven and directed
  marketing is measurement and accountability for
  actions. Decisions are based on facts, not
  opinions.
• Direct marketers build databases from facts,
  relying not so much on responses derived from
  survey, but more on conclusions derived from
  experimentation.

7
What Do Direct Marketers Test?

• Products and Services
• Media
• Offers/Propositions
• Copy Platforms
• Creative Formats
• Timing/Seasonality

8
TEST THE BIG THINGS
9
Sourcing Collecting Information

• Secondary Data
• Primary Data

10
(No Transcript)
11

• Primary data, collected via survey, can yield
  information about
• Behavior
• Intentions
• Knowledge
• Socioeconomic Factors
• Attitudes and Opinions
• Motivations
• Psychological Traits

12
(No Transcript)
13
How to Design An Experiment

• Control
• Randomization
• Statistically-valid sample size

14
How to Track Responses

• Key Codes

15
Response rate break-even analysis

• Control vs. experimental packages
• Direct marketers test or experiment with
  different offers and campaign themes to determine
  which one generates the greatest response rate

16
Promotion Cost ------------------------
Break-even Number of Sales Unit Profit per Sale
17
(No Transcript)
18
Samples Estimations

• Random sample designs
• Simple Random Samples
• Systematic Random Samples
• Stratified Random Samples
• Cluster Samples
• Replicated Samples
• Sequential Samples
• Determination of sample size

19
Lets first define the four terms confidence
level (z-value) limit of error, expected (or
actual) response rate and sample size -- that
enter into the calculation         Confidence
Level (Z-value) This is the value from a normal
distribution that corresponds to the chosen
confidence level. For example, the 90, 95, 99
confidence levels correspond to z-values of 1.65,
1.96, 2.58 respectively.         Limit of
Error The number of percentage points by which
the researcher is allowed to miscalculate the
actual response rate. A 20 percent limit of
error, assuming a 1 percent response rate, for
example, could result in a range of actual
response as low as 0.8 percent to as high as 1.2
percent or, 1 ? 20 of 1.       Expected
(Actual) Response Rate The number of times, in
percentage, that response is expected to
occur.       Sample Size The number of
observations in the experiment, or test. This
is, for example, the number of pieces mailed in a
test from which the response is to be determined.
20
To illustrate the use of the above formula, one
can determine the sample size required to be
mailed as a test when the expected response rate
is 1 the desired limit of error is ? 0.2 at a
confidence level of 95. Thus R 1 ...
0.01, expressed as a decimal 1 - R 99 ...
0.99, expressed as a decimal Z 1.96,
corresponding to a 95 confidence level E
0.2 ... 0.002, expressed as a decimal N
to be determined Substituting the above values
into the formula for the determination of sample
size, provides this solution N (0.01)
(0.99) (1.96)2 (0.002)2 (0.01
(0.99) (3.8416) (0.000004)
0.03803184 0.000004 9,508 pieces
to be mailed
21
(No Transcript)
22
Measurement of Differences

• Hypothesis testing
• Types of errors in hypothesis testing
• Statistical evaluation of differences

23
Hypothesis Testing Hypotheses are typically
stated in negative terms that is, a null
hypothesis (H0) versus an alternative hypothesis
(Ha) in a form such as the following H0 Direct
mail response from the test promotion is at or
below direct mail response from the control
promotion. Ha Direct mail response from the test
promotion is above direct mad response from the
control promotion. The null hypothesis, then,
states that direct mail response will not be
better than the control. Our measurement sets
out to disprove this null hypothesis.
24
Types of Errorin Hypothesis Testing

• Type One
• Results when the decision-maker rejects the null
  hypothesis even though it is, in fact, true ...
  i.e., taking an action when one shouldn't
• Type Two
• Results when the decision-maker accepts the null
  hypothesis when, in fact, not true ... i.e., not
  taking an action when one should.

25
Assume that a sample has been properly selected
and is of an
adequate size. Assume further that an experiment
has been
designed and implemented in a valid manner. It
now remains for
the direct marketer to be able to recognize the
difference in the
response rate from a test and that from a
control, with some
degree of confidence and within an acceptable
limit of error.
26
Test Control Totals Response
A C A C Non-response
B
D B D
Total mailed A B
C D A B C D N
The statistic ?2 is computed as follows ?2
N ? (A x D) - (C x B) ? - N/22 (AB) x
(CD) x (AC) x (BD) Here is a sample
calculation Test Control Totals
Response 200 100
300 Non-response
800 900 1700 Total mailed
1000 1000 2000 ?2 2,000
x ? 180,000 - 80,000? - 1,0002 1,000 x 1,000
x 300 x 1,700 ?2 38.4 ... which is significant
at the 99 level since it exceeds the critical
value in the X2 table for one degree of freedom
for a significance level of 0.001, given as 10.83
27
STRUCTURING and EVALUATING AN EXPERIMENT

• State the hypothesis
• Develop, by a priori analysis, the assumptions
  required and compute the appropriate sample size
• Structure and perform the experiment
• Develop, by a posteriori analysis, statistics for
  judging hypothesis validity
• Make the decision