M303 Data Editing, Coding, and Descriptive Analysis
Where we are
Fundamentals of Data Analysis
Preparation of data for analysis
Descriptive stats, simple and cross-tabs
Hypothesis testing
Correlation and regression analysis
Secondary data
Primary data
Observational and Qualitative Methods
Surveys
Sampling
Sampling schemes
Sample size
Problem
Formulation
Research
Design
Data
Collection
Data Analysis &
Recommendation
Preliminary stages of MR
Problem actionable management problem research objectives research questions hypotheses
Decide on the type of research
Exploratory, descriptive, and causal
Learning Objectives
You should be able to
Explain what data editing is
Learn how to code open- and closed-ended questions
Learn some useful tips in data coding
Learn the tools used in basic data analysis: descriptive stats, simple and cross tabulations
Learn the Web Tools implementation of each
Previous-class review
A confidence interval gives a plausible range for an unknown population parameter.
TReducing the allowable sampling error generally requires a larger sample.
TBar charts are useful for comparing frequencies across categories.
TA line chart is usually the clearest choice for showing change over time.
TWhat is data editing?
Data editing corresponds to identifying problems in the data such as
Interviewer errors
Omissions
Inconsistencies
Examples:
Omission of some unclear questions by most respondents
Some respondents might choose more than one alternative even though the question specifically asks them to choose only one.
How should we code the data?
How to code a response depends on whether the corresponding question is open or closed-ended:
Coding closed-ended questions involves specifying how to enter responses into the analysis software. We will use the instructor-designed Web Tools; Excel, SPSS, SAS, and R are other possible environments.
Generally, we assign numerical values to the response categories
Open-ended questions are relatively more difficult to code since every respondent might respond differently
A lengthy list of possible responses might be generated
How should open-ended responses be coded? Example 1
You show an email to your respondent first, and then ask the following question:
Please think back to the email you just read and answer the following questions:
What was the email about?
What was the main message?
How to code?
Correct answer?
No answer?
Wrong answer?
Kind of correct answer?
How should open-ended responses be coded? Example 2
How should closed-ended responses be coded? Likert scales
I think that Amazon.com provides better prices than BN.com.
Strongly Disagree · Disagree · Neutral · Agree · Strongly Agree
| Subject | Circled | Coded as |
|---|---|---|
| #1 | Disagree | 2 |
| #2 | Strongly Agree | 5 |
| #3 | Strongly Disagree | 1 |
Coding key: Strongly Disagree = 1 · Disagree = 2 · Neutral = 3 · Agree = 4 · Strongly Agree = 5
How should closed-ended responses be coded? Semantic differential scales
Please evaluate Amazon.com on the following dimensions.
Subject #1 marks: Modern 7 · Reliable 5 · Quality 4 · Price 5
Scale positions: 1 · 2 · 3 · 4 · 5 · 6 · 7
Index-variable coding
Subject · Modern · Reliable · Quality · Price
#1 · 7 · 5 · 4 · 5
#2 · 4 · 1 · 6 · 2
Subject #2 marks: Modern 4 · Reliable 1 · Quality 6 · Price 2
Example: coding rank-order responses
Please rank the importance of the following factors when you are choosing a restaurant.
| Factor | Subject #1 | Subject #2 | Subject #3 |
|---|---|---|---|
| Quality of the food | 1 | 1 | 2 |
| Atmosphere | 2 | 4 | 5 |
| Price | 4 | 2 | 1 |
| Proximity of the location | 5 | 3 | 3 |
| Reviews | 3 | 5 | 4 |
| Subject | Food | Atmosphere | Price | Location | Reviews |
|---|---|---|---|---|---|
| #1 | 1 | 2 | 4 | 5 | 3 |
| #2 | 1 | 4 | 2 | 3 | 5 |
| #3 | 2 | 5 | 1 | 3 | 4 |
How to interpret data from rank order scales?
Calculate the sum of frequency × rank for every option and compare
Ex: Please rank the importance of following factors in your choice of hospitals
Location _________ (1st - 15, 2nd – 20, 3rd – 35)
Reputation _________ (1st - 25, 2nd – 20, 3rd – 25)
Cleanliness _________ (1st - 30, 2nd – 30, 3rd – 10)
Location: (1 x 15) + (2 x 20) + (3 x 35) = 155
Reputation: (1 x 25) + (2 x 20) + (3 x 25) = 140
Cleanliness: (1 x 30) + (2 x 30) + (3 x 10) = 120
How should a check-all-that-apply response be coded?
| Subject | Circled | Index variables | ||||
|---|---|---|---|---|---|---|
| Snail | WoM | Advisor | Catalog | |||
| #1 | 2, 3, 4 | 0 | 1 | 1 | 1 | 0 |
| #2 | 5 | 0 | 0 | 0 | 0 | 1 |
| #3 | 1, 3, 4, 5 | 1 | 0 | 1 | 1 | 1 |
Example: coding a categorical response
Which is your favorite online bookstore? (circle one)
1. Amazon.com (A) · 2. BarnesAndNoble.com (B) · 3. Other (O)
When should you use each coding scheme?
Choice – if you want to calculate frequencies
ANOVA: Compare the mean of another variable (let’s say satisfaction) for the three groups (A, B, and O).
Index Variables – if you want to use the choice of A versus other choices (i.e., B and O are combined as choices other than A) in a regression.
Dummy Variables – if you want to use the three-fold choice (A, B, and O) in a regression.
When, how, and why should we use dummy coding?
When?
Nominal (categorical) variables with one possible answer (Ex: “check only one of the following”, “dichotomous”)
How?
For M categories, we need to create M − 1 dummy variables
Ex: Gender - Female & Male → one dummy
Ex: Race - White, Black, Hispanic, Asian & Other → four dummies
Why?
The only way to use nominal (categorical) variables in regression analyses
Dummy Coding Example
Original categories
| Subject | Race |
|---|---|
| John | White |
| William | Black |
| Michele | White |
| Jesus | Hispanic |
| Ali | Other |
| Yanwen | Asian |
| Taylor | Black |
Dummy-coded variables
| Subject | Asian | Black | Hispanic | White |
|---|---|---|---|---|
| John | 0 | 0 | 0 | 1 |
| William | 0 | 1 | 0 | 0 |
| Michele | 0 | 0 | 0 | 1 |
| Jesus | 0 | 0 | 1 | 0 |
| Ali | 0 | 0 | 0 | 0 |
| Yanwen | 1 | 0 | 0 | 0 |
| Taylor | 0 | 1 | 0 | 0 |
Other is the reference category: all four dummy variables equal 0.
What are some useful tips in data coding?
Create variables that are consistent with your study objectives.
Combine variables when necessary.
Example: Average responses to similar questions.
Please indicate how much you like Brand X (1–7).
I think Brand X is a good brand (1–7).
I enjoy what Brand X has to offer me (1–7).
Example: If very few respondents select −2 (very unsatisfied), it may be appropriate to combine −2 and −1 (unsatisfied) into a single −1 category.
Data Coding Exercise
Microsoft’s “I am a PC” ad:
http://www.youtube.com/watch?v=HrmF-mPLybw
Review the survey I distributed, then take 3 minutes to respond.
For the next 3–5 minutes, examine each question individually and write down how you would code your response.
Next, discuss how to code each question in groups of 3–4 for another 3–5 minutes. Agree on a coding approach for each response.
What are the tools used in basic data analysis? Descriptive Statistics
Descriptive statistics: statistics used to summarize the variables of interest
Includes:
Measures of central tendency such as mean, median and mode
Measures of dispersion such as range, standard deviation, and variance
Measures of shape such as skewness and kurtosis
What are the tools used in basic data analysis? Simple Tabulations
| Race | Frequency |
|---|---|
| White | 102 |
| Black | 89 |
| Hispanic | 75 |
| Asian | 26 |
| Other | 12 |
| Platform | Percentage used to check social media |
|---|---|
| Tablet | 39% |
| Desktop/Laptop | 20% |
| Smartphone | 41% |
| Diamond cut level | Average price ($) |
|---|---|
| Fair | 2,125 |
| Good | 2,752 |
| Ideal | 2,958 |
| Signature Ideal | 3,015 |
| Very Good | 3,450 |
| Gender | Average satisfaction (1–7) |
|---|---|
| Female | 4.8 |
| Male | 4.5 |
Simple tabulation tables — Example: Frequency Distributions
Reports the number of responses that each question received (or number of times each response category is chosen)
Organizes data into groups
Shows number of observations that fall into each group
Can be illustrated simply as a number or as a percentage
Response categories may be combined for many questions
Should result in categories with worthwhile number of responses
Simple tabulation tables — Example: Frequency Distributions
| Enrollment plans | Number | % |
|---|---|---|
| Yes, I would enroll | 124 | 11.1 |
| I would probably enroll | 211 | 18.9 |
| I am not sure now | 376 | 33.7 |
| I would probably not enroll | 204 | 18.3 |
| No, I would not enroll | 200 | 17.9 |
| Total | 1,115 | 99.9 |
| Combined categories | Number | % |
|---|---|---|
| Yes, I would enroll | 124 | 11.1 |
| I would probably enroll | 211 | 18.9 |
| Uncertain or uninterested | 780 | 69.9 |
| Total | 1,115 | 99.9 |
| Income | Number | % |
|---|---|---|
| Below $4,000 | 64 | 6.0 |
| $4,000–$6,999 | 122 | 11.4 |
| $7,000–$9,999 | 161 | 15.1 |
| $10,000–$12,999 | 110 | 10.3 |
| $13,000–$15,999 | 160 | 15.0 |
| $16,000–$20,000 | 327 | 30.7 |
| More than $20,000 | 123 | 11.5 |
| Total | 1,067 | 100.0 |
| Combined categories | Number | % |
|---|---|---|
| Less than $7,000 | 186 | 17.4 |
| $7,000–$15,999 | 431 | 40.4 |
| $16,000 and above | 450 | 42.2 |
| Total | 1,067 | 100.0 |
What are the tools used in basic data analysis? Cross Tabulations
Statistical analysis technique to study the relationships between variables
Sample is divided to learn how the variable of interest (also called the dependent variable) varies across different groups of another variable (also called the independent variable)
Ex: How income level (high versus low) affects the purchase likelihood (purchase versus no purchase) → DV: Purchase, IV: Income
The two variables (row and column, i.e., DV and IV) that are analyzed must be nominally scaled
In-Class Exercise
A diamond is forever
http://www.youtube.com/watch?v=8pU6WQXkiOU
http://www.youtube.com/watch?v=plg7Uw9NeDY
Near Monopoly: De Beers
Market share was 90% in early 1990s
"A Diamond is Forever"
Named the "Best advertising slogan in the twentieth century" by Advertising Age
In-Class Exercise (Continued)
"A Diamond is Forever"
Creates an emotional attachment
Relationship ↔ Diamond
Creates a highly durable good image (forever)
Life-long consumption → Higher willingness to pay
Prevents an effective secondary market (forever)
In-Class Exercise (Continued)
Online market
Blue Nile
Founded in 1999, Seattle
Largest online diamond retailer in the U.S.
4Cs
Carat
Cut
Color
Clarity
http://www.youtube.com/embed/qDNMkeWoYRI
In-class exercise: Web Tools illustration
Open diamondanalysis.xlsx in the instructor-designed Web Tools
9244 unique SKUs
Characteristics for each diamond 4Cs
Daily prices of each diamond
Calculate descriptive stats (such as mean, median, variance etc.) for the price
Create a frequency table for cut (cut levels versus count of diamonds in the corresponding cut level)
Create a simple tabulation of color versus average prices
Create a cross tabulation of color versus cut frequencies
Use the instructor-designed Web Tools to reproduce the descriptive statistics, frequency table, and cross-tabulation.
koraycosguner.github.io/tools.html ↗Use the instructor-designed Web Tools
Open the M303 Web Tools
Use the browser-based tools for descriptive statistics, frequency tables, cross-tabulations, hypothesis testing, regression, and conjoint analysis. Required data files and instructions will be posted on Canvas.
koraycosguner.github.io/tools.html ↗To sum up
Data editing
Data coding
Open-ended, closed-ended questions
Some tips in data coding
Tools that are used in basic data analysis and their Web Tools implementations
Descriptive stats
Simple tabulations
Cross tabulations