Conditional branching—direct skip to relevant question
Quality of response
Willingness to write out answers or respond to multiple closed-ended questions
Willingness to read and follow instructions is limited
Reliability and browser compatibility issues
Traditional surveys: Have you bought a new car during the last six months? If not, please skip to Question 11.
Conditional branching: Respondent will be taken to the appropriate question according to answer
Customization of questions
E.g., consumer lists three brands subsequent questions ask about these specific brands by name
Does it make a difference if subjects (i.e., consumers, people) are treated one way or another way in terms of some outcome (e.g., likelihood of purchase?)
Will consumers rate a red car as more exciting but a blue one as more reliable?
Are those who hear German (as opposed to French) music more likely to buy beer?
Will students who are told to expect an essay exam do better than those told to expect a multiple choice exam even if everyone ultimately takes a multiple choice exam?
Testing what people actually do rather than way they say or think they will do
Useful in trying to determine causation (e.g., does a product sell more if the packaging is red rather blue?)
Control for factors that are not equal in real life (e.g., If those who pay restaurant checks with credit cards are more likely to be reimbursed by their employers, it is not clear if the credit use was the cause of the higher tip)
Test and rule out competing explanations (subject to some caveats)—e.g., does texting while driving cause accidents because (1) eyes are removed from the road, (2) attention is diverted to the conversation, or (3) a combination
Two Basic Types of Experiments
Between-subject: Different groups of people are treated the same except for the variable or variables manipulated
E.g., One group shops in simulated store in which a credit card logo is displayed; the other group shops in the same simulated store but the credit card logo is removed
Within-subject: The same individual is treated differently at different times (e.g., at time 1 is given cola drink that includes vanilla and one without at time 2)
Subjects are usually counter-balanced to rule out order effects (1/2 receives treatment A first and then B, the other half receives B, then A)
Subjects in different groups are usually treated differently
E.g., for some, “target” product is given better shelf space
Everyone is treated the same way, so nothing is being manipulated for comparison
Ash’s Instant Coffee Study
GROCERY SHOPPING LIST
6 cups of yogurt
GROCERY SHOPPING LIST
6 cups of yogurt
Respondents were asked to describe their impressions of a housewife based only on her shopping list. These shopping lists differ only on one item.
Confound: An illusion of one variable causing another because of correlation between two variables
E.g., does having more toys cause children to be more intelligent?
Some types of confounds
A appears to cause B but it is actually B that causes A
Diversification and profitability. Diversified firms tend to have higher profit levels. This may be because more profitable firms need to find ways to re-invest profits rather than because diversification causes profitability
Sales of a brand of soda are higher during weeks of heavy advertising. However, advertising is allocated when the stakes are greatest (e.g., during holidays and summer months). Thus, anticipated sales actually cause advertising.
You will NOT be asked to draw these diagrams on the exam or to explain the specifics of the different types of confounds.
Some types of confounds
A appears to cause B but both A and B are actually caused by C
On the average, the more toys a child has, the higher his or her IQ. Both the number of toys and IQ may be caused by family resources such as income (providing for better nutrition and better education)
Individuals taking anti-depressive medication have higher rates of suicide than the general population. This is because those who take anti-depressive medication are more likely to be depressed. In fact, individuals suffering from depression are less likely to commit suicide than those with depression who do not take medication.
On the average, students who sit in the front of the class end up with higher grades. However, students choose where they sit.
Some types of confounds
A appears to cause B but both A is correlated with C, with C ultimately causing B
Sales of laundry detergent are higher during weeks when the brand is advertised. However, when the brand is advertised, there is always a price promotion (temporary price reduction). Most of the sales increase results from the price discount with a much smaller portion coming from advertising. The deeper the discount, the greater the sales.
Members of fraternities and sororities on the average have higher GPAs than people not in the Greek system. There may be some benefits to Greek membership causing higher grades, but members of the Greek system also tend to come from more privileged backgrounds.
What is cause, what is effect, and what is coincidence?
Correlation is not necessarily cause
“Lurking” factors may be real cause
Does sitting in front of the room cause higher grades?
Does Prozac cause suicide?
Do fish-heavy diets cause stomach cancer?
Does fraternity/sorority membership cause higher grades?
Confounds: The Bottom Line
Correlation ǂ cause!
One can often make a plausible—even compelling—hypothesis of causation based on the observed relationship (but a hypothesis is still an open question)
The causation may be the other way around (B causes A rather than A causes B)
The variable that seems to cause the other may be correlated with the actual cause (A does not cause C, but a is correlated with B and B causes C)
Experiments can be used to test for causality (with certain caveats) by controlling for potentially confounding variables
Some other influences on the relationship between variables
Restricted range: One population includes only a portion of another or the sample includes only a small portion of the population range—e.g.,
High school GPA school is a limited predictor of college GPA since students with higher high school GPAs tend to get into more competitive colleges
Among residents of Rancho Palos Verdes, household income does not predict dental health. In society as a whole, many people have difficulty paying for quality dental care, but most residents of Rancho Palos Verdes are relatively affluent
“Diluting” variables: Variables not considered “drown out” much of the effect of the variable of interest—e.g.,
Income is a limited predictor of spending. Although a certain level of income (or wealth) is needed to buy certain things (necessary condition), income is not a sufficient condition. Many high income individuals are rather stingy and many lower income individuals live above their means.
Ethnicity is a limited predictor of food preferences. Many Caucasians, African-Americans, Latinos, and Native Americans like raw fish.
Looking at consumes in the field—e.g.,
Searching for product category area
Number of products inspected and time spent on each
Apparent scrutiny of labels or other information
Involvement of others
Behavior under limiting circumstances (e.g., time constraints)
LOW TO HIGH
(DEPENDING ON CODING AND
Not experiments unless
Two or more groups of people are treated differently (e.g., get different food version) or
The same person is being treated differently at separate times (e.g., half the participants receive new formulation, then current; half the participants receive in the opposite order)
Each respondent is given three items: One current, one new, and one duplicate of either old or new
Asked to identify the one that is different and explain why
Groups of 5-12 consumers assembled
Start out talking generally about context of product
Gradually “focus” in on actual product
Usually NOT the best approach. Should NOT be chosen as default research method!
FOCUS GROUP COSTS:
(ESPECIALLY FOR THE AMOUNT
OF INFORMATION COLLECTED)
Focus Groups: Potential Uses
Identifying possible issues of concern with a new product
Probing complex issues where different factors and issues may affect opinions
Probing differences in perspectives among different groups
Will consumers pay $500.00 to reduce the weight of a laptop computer by 2 lbs?
Conjoint Analysis: Disadvantages
May be difficult for subjects to rate many combinations
May need a large number of subjects for accurate measurement/sufficient precision
Must identify relevant attributes and levels in advance
Subject must know about product category (attributes must be meaningful)
Types of scanner data
Supermarket club. This includes purchases by the specific customer when shopping at the respective chain (assuming that the customer presents his or her card). Purchases at other locations are not counted. Demographic information may be of limited accuracy. Shoppers are often motivated to join by large discounts (often 20-30%). This is also a method of price discrimination. Members may be given individualized coupons for possible products of interest. Only available for grocery products.
Scanner data panels. In some communities, people can sign up to be part of a “panel.” Purchases at all local retailers are included (e.g. supermarkets, gas stations, drug stores, convenience stores). For a given customer, the database also includes TV viewing and demographics. Only available for grocery products.
Aggregated retail sales records. Records of sales volumes of products by UPC may be available from an assortment of retailers. This information is NOT tied to individual customer data and purchase history. This is available for more types of products.