Market Research

1
Market Research

• Tuesday, February 8th

2

• This class
• Experiments and t-tests
• Analyze some data
• Midterm question examples will be posted
• Debate teams will be posted
• Next class
• Midterm review quiz
• General questions

3
Debate Topics

• Marketing to children
• Marketing of harmful goods
• Marketing promotes materialism
• Price discrimination
• Marketing creates unnecessary needs
• Marketing adjusts perceptions of ideal body types
• The use of low-cost labour for competitive
  advantage
• Marketers promote meaningless attributes to help
  with differentiation
• Marketers should provide full information about
  their products and practices

4
Experiments Defined

• What is an experiment?
• Surveys, observational research, and even focus
  groups involve the assembly of existing data
• In experiments, the researcher actually changes
  things

5
Experiments Defined

• An experiment is the form of research used to
  establish causality
• It involves manipulating (controlling) the level
  of one variable and observing the response in
  another

6
Experiments Variables

• Any characteristic on which observational units
  (e.g. people, firms, etc.) differ
• Observable (concrete, manifest) or unobservable
  (conceptual, abstract, latent)
• Categorical (discrete), e.g. religion
• Dichotomous, e.g. gender
• Continuous, e.g. attitude, age, income

7
Experiment Variables

• Independent Variable (IV)
• A variable whose value is systematically varied
  by the experimenter
• Dependent Variable (DV)
• A variable that is assumed to be affected by the
  IV
• Mediator variables (intervening) are one class of
  DVs

8
Experiments Variables

• Mediator (Intervening)
• A variable that is affected by the IV that in
  turn affects the DV
• Moderator
• A variable that defines the scope of the
  relationship between the IV and DV
• Extraneous
• Any other variable that may affect the
  relationship being studied

9
Experiments Types

• Laboratory experiments
• Can control extraneous causal factors
• Minimize unwanted influences
• Field experiments
• Realistic setting
• Control some aspects of situation but not all
• e.g. Mars offered different candy bars to
  different markets

10
Experiments Causality

• Experiments are designed to demonstrate that a
  change in one variable causes a change in another
• What does A causes B mean?
• A is one cause of B
• A makes occurrence of B more probable
• Relationship is inferred, never certain
• How do we establish causality?

11
Experiments Causality

• Concomitant variation (correlation)
• A and B vary together
• e.g. advertising and sales, price and sales, etc.
• Temporal antecedence
• A occurs before B
• e.g. ? advertising occurs before ? sales
• No plausible alternative explanation
• Known as internal validity
• ? B not caused by ? C, ? D, ? E, etc.
• e.g. ? sales not caused by ? price of a substitute

12
Note Correlations

• Whats the difference between correlation and
  covariation?
• Both are measures of the extent to which two
  variables change together
• Covariation is the average of the cross-product
  of the deviation scores (not bounded)
• Correlation is the same, but of the standardized
  scores (bounded by -1 and 1)

13
(No Transcript)
14
ExperimentsCorrelation-Causality Fallacy

• Correlation alone does not imply causality
• Children who watch violent movies appear to be
  more violent
• Do violent movies increase childrens violence?
• Implication do not let children watch violent
  movies
• There are more fire trucks at larger fires
• Do fire trucks cause larger fires?
• Low income groups are less intelligent
• Does low income cause lower intelligence? Does
  lower intelligence cause lower income?

15
Experiments Validity

• Experiments that rule out competing explanations
  are said to have a high degree of internal
  validity
• What might threaten the internal validity of an
  experiment?

16
Internal Validity Threats

• History
• Influence of extraneous variables during
  experiment
• e.g. Ragu changed advertising when Campbells
  tested Prego
• Maturation
• Changes in individuals over time
• Testing
• Testing or observing individuals may change their
  behaviour
• Instrument decay
• Changes that affect measuring technique

17
Internal Validity Threats

• Selection bias
• Differences may exist in groups ahead of time
  without randomization
• Mortality
• Loss of individuals over course of experiment
• Regression towards the mean
• Extreme individuals tend to regress towards mean
• Interactive effects
• Any of the above threats may be stronger within
  treatment levels

18
Experiments Designs

• System of notation
• Xi Experimental treatment i
• The variable whose effects we want to measure
• e.g. different prices, package designs, ad.s,
  etc.
• Oi Observation of test units i
• Involves taking measurements of individuals,
  groups, or entities whose response to the
  experimental treatment is being tested

19
Pre-Experiments Designs
20
Pre-Experiments Designs

• One-shot design
• Expose individuals to treatment variable and then
  measure dependent variable
• No comparison either beforehand or to another
  group
• No basis for a judgment of causality

21
Pre-Experiments Designs

• One-group pretest-posttest design
• Measure DV before and after treatment
• Vulnerable to history, maturation, testing,
  instrument decay, regression towards mean,
  mortality

22
Pre-Experiments Designs

• Static-group comparison design
• Measure DV in two groups, one of which has been
  exposed to treatment
• Comparison to different group helps remove
  history, maturation, testing, instrument decay,
  regression, and mortality effects
• Vulnerable to selection bias and interaction
  effects

23
Experiments Designs
24
Experiments Designs

• Posttest control group design
• Static group comparison inc. randomized groups
• Removes effects of selection
• Cannot measure testing or interaction effects
• Effect of treatment O2 - O1

25
Experiments Designs

• Pretest-posttest control group
• Measure DV for two randomized groups, apply
  treatment to one and then measure DV again
• Can measure effects of testing
• Cannot measure interaction effects
• Treatment effect (O2 O1) (O4 O3)

26
Experiments Designs
27
Experiments Designs

• Solomon four-group design
• Removes threats to internal validity
• Treatment effect (O2 O1) (O4 O3)
• Treatment w/o testing O5 O6
• Treatment with testing O2 O4
• Can measure interaction (O2 O4) (O5 O6)

28
Experiments Validity

• Experiments are used to maximize internal
  validity
• However, should also consider external validity
• External validity refers to the extent to which
  we can generalize to the population
• Depends on extent to which sample and setting are
  representative

29
Research Samples

• Ideally we want to be able to generalize to the
  population from which our sample came
• Random sample each member of population has an
  EQUAL probability of being selected
• Stratified random sample each member of a
  specific group has an equal probability of being
  selected
• Convenience sample
• Judgment sample

30
External Validity Threats

• Selection ( mortality)
• Typically university students
• Unformed sense of self
• Uncrystalized attitudes
• Stronger need for peer approval
• Adept cognitive skills
• Less experienced consumers
• Cash and time poor
• Setting
• Demand effects (testing effects)
• Effects of being in lab
• Cognitive mindset (feeling of being tested)
• Compliance and / or suspicion
• Low motivation
• Simplified information

31
Football

• You may have noticed that when the season
  starts, the weather is always warm and summery.
  But after the fullbacks, halfbacks, quarterbacks
  and any other fraction backs have thundered from
  goalpost to goalpost a few times, nature gets
  upset, and cold weather begins. You may also
  notice that countries where football isnt
  played, such as Mexico, Jamaica and Egypt, do not
  have cold weather. In Canada, on the other hand,
  where they have one additional player on each
  side, winters are worse than here. Small
  wonder, Benjamin concludes, that during the NFL
  strike we had freakishly mild weather across much
  of the northern U.S.
• Alan L. Benjamin in a Letter to the Chicago
  Tribune

32
Questions

• What is the independent variable?
• What is the dependent variable?
• What are the mediating variables?
• What elements of causality do we have?

33
Questions

• What is the hypothesis?
• What is the theory?
• What is problem?

34
Experiments Hypotheses

• A statement that describes a relationship between
  two variables
• Causal or correlational
• Should be testable
• Should be better than its rivals
• Occams Razor (Ockhams) simpler
• Greater range

35
Experiments Theories

• A set of systematically interrelated concepts,
  definitions and propositions that are advanced to
  explain and predict phenomena
• Theories tend to be abstract and involve multiple
  variables
• Hypotheses tend to be simple, two variable
  propositions involving concrete instances
• Hypotheses flow from the theory

36

• How would we test the hypothesis that football
  causes winter?

37
Example

• Research question
• Do media images affect individuals perceptions
  of their ideal body type?
• Hypothesis
• Advertisements showing overly slim females and
  males cause consumers to attempt to attain or
  desire such body forms
• What moderator variables could there be?
• Do we expect this relationship to be stronger for
  males or females? Age? ?

38
Example

• Possible mediator variables
• Self-esteem, clothing tastes, etc.
• Alternative hypotheses
• Individuals are already concerned with body
  image, but changes in clothing taste have made
  appearance more important
• Direction of causality
• Experiment to examine hypothesis

39
Features of Experiments

• Random assignment
• Controls for extraneous variables
• Measure other extraneous variables
• Including possible mediators and moderators
• External validity (generalizability)
• Extent to which findings are generalizable to
  other populations and settings
• Random sampling

40
Statistical Methods

• Descriptive versus Inferential
• Descriptive
• Mean, median, mode
• Variance, standard deviation
• Correlation, covariance, regression
• Inferential
• T-test, ANOVA, chi square, etc

41
Statistical Methods T-Test

• Differences between means
• e.g. Are men taller than women? Are short people
  happier than tall people?
• Group is categorical variable (only 2 groups for
  t-test)
• e.g gender, religion, tall vs. short, etc.
• Variable of interest is continuous
• e.g. height, happiness, age, income, etc.

42
Data Analysis

• Are men taller than women?
• Mean height of men 178 cm
• Mean height of women 165 cm

43
Females
Males
44
Data Analysis

• How do we know whether the difference in the
  samples reflects a difference in the actual
  populations?
• That is, do the two populations (males and
  females) actually differ along this dimension
  (height)?
• By making some assumptions we can calculate the
  probability that the difference we observed would
  occur given there was in actual fact no
  difference in height

45
Data Analysis

• How do we calculate this probability?
• We can calculate the number of standard
  deviations the difference is above the mean
  difference
• If we know what this distribution looks like we
  determine the probability that the difference
  came from that particular distribution

46
Analysis Example

• What is the probability that someone has an IQ
  greater than 130?
• If we know what the distribution of IQs looks
  like we can calculate the probability
• We need to know the mean, standard deviation, and
  shape of the distribution
• In this case µ 100, s 15, and normal

47
Analysis Example

• 130 is 2 standard deviations above the mean
• (130-100)/15 2
• This is known as the z-score
• This formula tells us how many standard
  deviations our data point is from the mean

48
Analysis Example

• Under a normal distribution
• 68 of points lie within 1 standard deviation of
  the mean
• 96 of points lie within 2 standard deviations

2s
1s
49
Analysis Example

• What is the probability that someone has an IQ
  greater than 130?
• p (1 - .96)/2
• p .02 or a 2 chance
• i.e. 2 of the population has an IQ 130

50
Data Analysis

• Rather than looking at the difference between an
  individual data point and the mean, we are
  looking at the difference between a mean
  difference and the mean of the mean differences
• If we assume there are no differences, then the
  mean of the mean differences is zero
• Thus, we want to calculate how many standard
  deviations our mean difference is from zero
• To do this we need to know the standard deviation
  of the mean differences

51
Data Analysis
52
T-Test Formula
Because we dont know the actual standard
deviation of the mean differences, the statistic
is no longer normally distributed
The new statistic has a Students t-distribution
53
T-Tests

• The denominator is the standard deviation of the
  mean differences
• Imagine we took many samples (of men women)
• We could calculate the mean differences between
  each pair of samples
• Each mean difference will be different

54
T-Tests

• What would they look like if they were plotted?
• Normally distributed
• What would the variance be in this new
  distribution relative to the variances within
  each sample?
• The variance of the mean differences would be
  less than the variance in each sample
• This variance is known as the standard error of
  the mean difference

55
T-Tests

• However, we dont know the standard error
• So we estimate it from the information we do have
  (the standard deviations of the samples)
• The standard error is the s.d. divided by the
  square root of the sample size

56
T-Tests

• But first we must calculate the s.d.
• We assume that the s.d. in each population (i.e.
  men and women) is identical
• So it is more efficient to pool them into a
  single estimate of the s.d.
• We pool them into one s.d.

57
T-Test Variances
Generally, the variance is calculate as follows
We can pool the variances from two samples
58
T-test Standard Error

• Now we have a single estimate of the s.d.s in the
  population, we can calculate the standard error
  of the mean differences

59
T-tests

• Now we have everything we need to calculate the
  t-statistic
• This tells us how many s.d.s our mean difference
  is away from the mean in the t-distribution
• All that remains is to calculate the probability
  this could have occurred by chance alone

60
T-Test Procedure

• What t-distribution must we look at?
• Depends on the degrees of freedom, v (nu)
• v (n1 -1) (n2 1) n1 n2 2
• We must then find the critical t-statistic
• This is the minimum t-statistic that would be
  necessary for us to infer that our mean
  difference did not occur by chance alone
• Scientific standards dictate that there can be no
  more than a 5 chance of this

61
T-tests Critical t-values

• Here are the critical ts for a range of degrees
  of freedom
• v 5, t 2.571
• v 10, t 2.228
• v 20, t 2.086
• v 100, t 1.984
• Note that we do not distinguish whether our
  t-statistic is above or below the mean, unless we
  have a strong reason to predict this a priori

62
T-tests Critical t-values

• T-distribution for v 20
• Critical t 2.086

95 of points are between
2.086
-2.086
97.5 of points are below this level
63
Statistical Methods T-Test

• Hypothesis being tested
• H0 (Null Hypothesis) Means are equal in
  population
• We either accept or reject H0
• If we reject H0, we implicitly accept H1
• H1 (Alternative Hypothesis) Means are different
  in population

64
Statistical Assumptions

• Normality the data in each population is
  normally distributed
• Homogeneity of variance the two population
  variances are identical
• Independence of observations data within or
  between the two groups are not associated in any
  way

65
Data Analysis

• Are men taller than women?
• Mean height of men 178 cm
• (s.d. 8.62, n 47)
• Mean height of women 165 cm
• (s.d 7.65, n 38)
• Standard error of mean difference 1.79
• v 83, critical t80 1.99

66
Data Analysis

• Are men happier than women?
• Mean happiness men 1.91
• (s.d. 1.10, n 47)
• Mean happiness women 1.58
• (s.d. .79, n 38)
• Standard error of mean difference .21
• v 83, critical t80 1.99

67
Data Analysis

• Did question order matter?
• Mean happiness first 1.79
• (s.d. .83, n 43)
• Mean happiness second 1.74
• (s.d. 1.13, n 42)
• Standard error of mean difference .21
• v 83, critical t80 1.99

68
Data Analysis

• Are taller people happier than shorter people?
• Mean happiness taller 2.02
• (s.d. .67, n 43)
• Mean happiness shorter 1.50
• (s.d. 1.17, n 42)
• Standard error of mean difference .21
• v 83, critical t80 1.99