Notes

Introduction to STATA and Data Setup One of the most used software for data analysis in social science. Both User-friendly and Coding Options. A number of statistical methods are available. STATA is a software, not a method of analysis.

Data Inspection You can put the code in "command" or make the do-file. Use "summarize (or sum) variable name" to check the summary statistics such as # of observations, mean, standard deviation, min, or max. Use "det" option to return the detailed statistics. To see the histogram of a variable, use "hist variable name". "tab" & "tab2" : Frequency.

Data Preparation Some variables need to be recorded:

• Grouping
• Reverse scale

Some variables need to be combined into one variable:

• Multi-item scales: Loyalty1, Loyalty2, Loyalty3

Data Preparation - Cronbach Alpha

• Three different loyalty measurement items: Loyalty1, Loyalty2, Loyalty3.
• How do we generate one integrated measure for Customer Loyalty?
• Are our measurement items reliable?
• When to Use it? In survey research, when we have multiple items to measure one variable, we need to check the consistency among multiple items.
• Why? Usually, we use mean of multiple items (think about the case of customer loyalty). Thus, the internal consistency matters.
• What does it suggest? If the Cronbach alpha is sufficiently high (i.e., higher than 0.6), then we consider there is high internal consistency.

Data Preparation - Inter-Item Reliability

• Do the items really measure the same construct?
• Cronbach's alpha (1951).
• Formula:

• Scores over 0.6 are considered to have a high internal consistency
  • r: average inter-item correlation
  • N: number of items

Data Preparation - What if Cronbach's Alpha < 0.6?

• Leave out an item
• Check again
• If necessary, leave out another item, etc.

Data Analysis - Regression and Model Selection

Data Analysis - What Factors Are Important?

Regression Diagnostics - Multicollinearity

• One assumption of OLS is that the independent variables are independent of each other.
• What if this assumption is violated?
• If this assumption is violated, we deal with multicollinearity.
• One independent variable in a multiple regression model can be explained by the other independent variables in the model with the high degree of accuracy.
• Why does it matter?
  • Coefficients depend on other variables in the model.
  • The precision of the coefficients can decrease.
  • Significance of the coefficient can change.

Example

• To attract more visitors, a cinema decides to lower the price of a ticket and distribute flyers in some weeks.
• Now, the management wants to assess the effectiveness of the price discounts and flyers.

Regression Diagnostics - How To Detect Multicollinearity?

• Correlation Matrix
  • Relationship between:
    • X1 and X2
    • X1 and X3
    • X2 and X3
    • Etc.
• Very High Correlations indicate multicollinearity
  • Cutoff arbitrary, e.g., 0.9.

Regression Diagnostics - Correlation Matrix

Regression Diagnostics - Better Approach?

• Not only bivariate relationships matter (e.g., X1 and X2)
• But also it is a problem is one IV is a linear function of other IVs (X1 is a function of X2, X3 and X4).

Variance Inflation Factors (VIFs) What if VIF > 10?

• Leave out one variable (Effect captured by other variable(s), e.g., flyer effect incorporated in price effect).
• Combine variables in one variable
  • Sum of competitors' advertising expenditures instead of a variable for each competitor.
  • Factor analysis.

Marketing and Consumer Choice Consumer choice model in marketing has been improved in many ways. What marketing scholars have found:

• Consumer choice can be affected by sales promotion. In this case, brand switching, timing acceleration, quantity acceleration, stockpiling prevails.
• Consumer choice can differ a lot due to preference heterogeneity. Marketing scholars start estimating individual consumer choice using Bayesian modeling.
• Consumers are learning and make a decision strategically. For example, consumers learn a about the timing of sales promotion and delay the purchase till the anticipated sales promotion.

In marketing, there are many cases where consumers make a choice.

• Do consumers buy the product?
• Do consumers click the advertising?
• Do consumers respond to marketing actions?
• Do consumers like our brand?
• Do consumers follow us in social media?

Thus, in marketing, we have many cases where the dependent variable is a binary variable: 1 or 0. Then, How Do We Model Consumer Choice Behavior? Regression Model? Suppose we set up the following linear regression model:

Where $Choice_i$ = 1 or 0 and $X_i$ is an independent variable for consumer i.

Let's Do It in STATA Let's run the following model: $Social Media_i = β_0 + β_1 R_i + β_2 F_i + β_3 M_i + β_4 Age_i + β_5 Gender_i + ε_i$

Nothing Happens? What If We Predict the DV?

In Linear Regression Model...

Is the relationship between $X_i$ and $Choice_i$ linear?

Then, How Do We Set Up a Model? Logistic Regression - Modeling Consumer Choice When the DV is binary, we can use logistic regression. To do so, we need to understand "odd". Odd 0 p(occurring) / p(not occurring) = p / (1 - p) where p = the probability of choice. Logistic regression:

Let's estimate this logistic regression model in STATA! Find out which customers are most likely to follow us on social media.

$$Social Media_i = z_i = ln[p/(1-p)] = β_0 + β_1 R_i + β_2 F_i + β_3 M_i + β_4 Age_i + β_5 Gender_i + ε_i$$

Responses in our case is whether a customer follows us on social media.

Model Significance and Fit How Do We Evaluate the Model Fit?

• After running the model, we can read LR chi2 (df) and check the p-value.
• Pseudo R2: this is analogous to R2 and we should not interpret this statistic as the proportion of the variance of the DV explained by the independent variables.
• How to check the predictability of the model in STATA? estat classification.

Hit Rate

Coefficients Interpretation

• Note that we use logistic regression when the DV is binary and the model uses "log(p/(1-p))" when estimating the coefficients.
• Thus, the interpretation of the coefficient shouldn't be the same as that in linear regression.
• Interpretation: If one unit of X increases, log-odds increases or decreases.

Predicting the Probability in STATA. After running logistic regression, you can use "predict variable name, pr" to generate the predicted probability for each observation.

Consumer Decision Making Process Marketing Research and Consumer Behavior - Netflix Case

• More than 60,000 titles
• 154 million subscribers worldwide
• Watching 3.8 billion hours every month
• Revenue: US$ 15.8 billion
• Income: US$ 1.6 billion.
• Big Data Insights: Personalized Recommendations
• How accurate are Netflix's recommendations? 75% of viewing activity results from these suggestion.

Not Marketing Analytics but Any Other Perspective? Consumer Behavior?

Marketing Research and Consumer Behavior - Model of Consumer Behavior

• Marketing = understanding consumers
• Central question for marketers: How do consumers respond to various marketing efforts the company might use?

Consumer Decision Making Process

Need Cognition

• The process starts with need recognition.
• Consumers recognize a problem or needs.
• Internal stimuli (e.g., hunger or thirst).
• External stimuli (e.g., advertisement or chatting).

In this stage, marketers should research:

• What kind of needs or problems?
• What brings them?

Information Search

• Once consumers recognize needs, they are usually motivated to search for more information on products/services.
• Consumers obtain information from various sources:
  • Personal sources
  • Commercial sources
  • Public sources
  • Experiential sources

This stage is important because consumers create "consideration sets".

Evaluation of Alternatives

• Once consumers select brands for the consideration set (i.e., a set of final brand choices), they evaluate alternatives only in the consideration set.
• Careful calculation and logical thinking vs. little or no evaluation.
• In this stage, consumers evaluate products based on several attributes with difference importance!

Purchase Decision

• Form purchase intention and buy the most preferred brand.

Post-Purchase Behavior

• Both satisfied and dissatisfied consumers engage in the post-purchase behavior.
• What determines customer satisfaction: Expectation and Perceived Performance.
• The gap between expectation and perceived performance determines satisfaction.
• Marketers should promise only what their brands can deliver!
• Cognitive Dissonance: Discomfort caused by post-purchase:
  • Avoid acquiring drawbacks of the product they buy.
  • Avoid losing the benefits of products they do not buy.

• Marketers?
  • Should measure customer satisfaction regularly.
  • Set up a system to encourage dissatisfied consumers to complain and share problems.

Why don't dissatisfied consumers complain?

• They think it is not worth the time and effort.
• They think no one would be concerned about their problem or solve it.
• They do not know where to go or what to do.

What are the Implications for Marketers?

• By studying the overall buyer decision process, marketers may be able to find ways to help consumer move through it.
• E.g., Consumers do not buy a new product because they do not perceive a need for it, marketers can launch advertising messages that trigger the need and show how the product solves customers' problems.
• Marketers change consumers' unfavourable attitudes about their brands.

Implications for Modeling Consumer Choice? Do consumer Always Follow This Decision Making Process?

What is Consumer Involvement?

• Consumer involvement: the degree of information processing and the amount of importance a consumer attaches to a product while purchasing it.
• Consumers are highly involved in the decision making process when the product is expensive, risky, purchased infrequently, and highly self-expressive.
• Consumer involvement is low when a product is relatively inexpensive and frequently purchased.

Types of Consumer Decision Making Behavior

Consumer Decision Making Process There are Different Lens! Cognitive vs Emotional Decision Making

• In many cases, consumer decision making is cognitive (Economically optimal and rational choice).
  • Deliberative and careful
  • Information-based processing of product attributes
  • E.g., buying a house, choosing a university, buying an insurance.
• But some decisions are driven by emotion, subjective liking or feeling.
  • Subjective liking
  • Feeling
  • E.g., a person living in Bali or Singapore or Bangkok buys and wears a wool-lined leather jacket.

Many purchases have both components (cognitive and emotional).

• Cognitive:
  • Price of the phone
  • Any bundle?
  • Cost of service plan
  • Warranty
  • Performance (e.g., battery or storage)
• Emotional:
  • Considered "cool"
  • Favorite color (e.g., rose gold, metal black, titanium gray)
  • A personalized unique case

Why Should Marketers Determine It?

• Marketing strategies vary and depend on the decision making process.
• Cognitive decision making is often slow, more systematic, and exhaustive than emotional decision making.
• Emotional decision making is quicker and more spontaneous.

Different Marketing Strategies!

• Product Placement? Where to place the product?
• Advertising? What to emphasize?

Modeling Issue?

• Which variables should we include in the model to explain consumer choice.