Marketing Research Lecture 7: Data Collection III - Sampling PDF

Marketing Research Lecture 7: Data Collection III – Sampling PROFESSOR JIA LIU  From Last Class Problem formulation Data collection Research design Measurements...

Marketing Research Lecture 7: Data Collection III – Sampling PROFESSOR JIA LIU  From Last Class Problem formulation Data collection Research design Measurements Data collection Questionnaire Analysis and interpretation Sampling Reporting results HKUST, MARKETING RESEARCH  Outline of Today 1. Definitions of population and sample 2. Sampling procedures 3. Sample size and sampling error HKUST, MARKETING RESEARCH  POPULATION AND SAMPLING HKUST, MARKETING RESEARCH  Population Population All individuals or objects that meet certain requirements for membership in the group Examples All households in Hong Kong All HKUST students All cars sold in 2016 All transactions on HKTVmall in 2021 HKUST, MARKETING RESEARCH  Population and Census Population Census Information about every member of the population is collected. Examples: - Hong Kong Census and Statistics Department collects demographic information of the whole population in Hong Kong - HKTVmall collects information on every transaction in 2021 - Foodpanda collects all deliveries ordered by its users in 2020 HKUST, MARKETING RESEARCH  Population and Sample Population Sample Only a subset of the population is studied Objective: use the sample to draw conclusions about the target population 1. When the population data is not available 2. Or it is too costly to analyze the entire population HKUST, MARKETING RESEARCH  Population and Sample Population Sample Only a subset of the population is studied. Examples: - Average number of orders placed by customers on Taobao Parameter Statistic - Evaluation of Characteristics of the Characteristics of the shopping experience population sample on HKTVmall - Attitudes toward ad Sampling Error = parameter - statistic copies HKUST, MARKETING RESEARCH  Sampling Procedure 1. Define the target population 2. Identify the sampling frame channels to reach out to do the study, e.g. email 3. Select a sampling procedure which subset 4. Determine sample size 5. Select sample elements 6. Collect the data Question: How do you define the target population (Step 1)? HKUST, MARKETING RESEARCH  Sampling Frame Examples A list of population elements from which you draw the sample E.g., geographic areas, institutions, individuals, or other units Research problem Population A Sample Frame Attitude towards sports All readers of Mail addresses of pages in SCMP SCMP subscribers to SCMP Attitude towards HK All HK residents Telephone directory government of Hong Kong Customer shopping All Donki’s target ? experience at Donki customers Student satisfaction with All HKUST UG ? HKUST UG programs students HKUST, MARKETING RESEARCH  SAMPLE PROCEDURES HKUST, MARKETING RESEARCH  Sampling Procedure More control 1. Non-Probability ◦ Subjective procedure in which some members of the population have a zero or unknown probability of being selected 2. Probability ◦ Every element has a known and non-zero probability of inclusion in the sample Advantages of Probability Samples ◦ Allows quantification of sampling error ◦ Generally more representative HKUST, MARKETING RESEARCH  Non-Probability Samples 1. Convenience sample Being included in the sample is a matter of convenience Main problem is not knowing its representativeness based on 2. Judgment sample criteria Hand-picked sampling elements ask them to extend to send to other people match the criteria Most effective in situations where there are only a restricted number of people in a population who own qualities that a researcher expects from the target population. Snowball sampling (for special populations) 3. Quota sample Build a sample that looks like the larger population in certain dimensions HKUST, MARKETING RESEARCH  Example Convenience sample All subscribers Judgement sample All subscribes whose children are between age 5 and 20 criteria HKUST, MARKETING RESEARCH  Example of Quota Sample The population is divided into subgroups based on specific characteristics HKUST, MARKETING RESEARCH  Probability Designs 1. Simple random sampling 2. Stratified sampling 3. Cluster sampling 4. Systematic sampling 5. Multi-stage sampling HKUST, MARKETING RESEARCH  Simple Random Sampling Each element of the population has a known and Advantages equal probability of of being selected for the sample Easy to implement Disadvantages Mechanics Less efficient than stratified Table of random numbers sampling Computer generated random Usually more expensive than numbers cluster sampling Random digit dialing HKUST, MARKETING RESEARCH  Simple Random Sampling 1 2 3 4 5 6 7 8 9 10 11 12 Population Sample HKUST, MARKETING RESEARCH  Example: Opinion Polls HKUST, MARKETING RESEARCH  Stratified Sampling Two-Step Procedure 1. Divide the population into mutually exclusive and exhaustive subgroups (strata) 2. Randomly select a sample from each subgroup (strata) tedious procedure Feature Homogeneous within group Heterogeneous across groups HKUST, MARKETING RESEARCH  Stratified Sampling Population Sample HKUST, MARKETING RESEARCH  Example of Stratified Sampling Do Chinese consumers have the habit of drinking coffee? Groups: different age groups 1. Blow 20 2. 20 – 29 3. 30 – 39 4. 40 – 49 5. 50 – 59 6. Above 60 Randomly select 100 people from each group HKUST, MARKETING RESEARCH  Stratified Sampling Advantages over simple random sampling Efficiency: reduce the costs of drawing similar elements Effectiveness: ensure representation from each subgroup Disadvantages Identifying strata is costly and difficult Require larger sample with larger number of strata HKUST, MARKETING RESEARCH  Cluster Sampling Three-Step Procedure 1. Divide the population into mutually exclusive and exhaustive subgroups 2. Then randomly select one or more subgroups 3. Finally select either all (one-stage cluster sampling) or a random sample of the elements (two-stage cluster sampling) from the chosen subgroups Feature: large heterogeneous within group HKUST, MARKETING RESEARCH  Cluster Sampling Population Sample HKUST, MARKETING RESEARCH  Example of Cluster Sampling Do Chinese consumers have the habit of drinking coffee? Groups: 18 different districts 1. Island – Central & western 2. Island – Eastern 3. Island – Southern 4. Island – Wan Chai 5. Kowloon – Kowloon city 6. … Randomly select 5 districts Randomly select 100 from each selected district HKUST, MARKETING RESEARCH  Systematic Sampling Procedure: 1. Determine the sample size N 2. Divide the population of size M by N, and let d = M/N 3. Pick a random start (key for randomness → every one has the equal chance of being selected) 4. Select every d elements Features: Systematically spread the sample through the population list Efficiency depends on the ordering of the list HKUST, MARKETING RESEARCH  Systematic vs. Simple Random Sampling Video HKUST, MARKETING RESEARCH  Multi-Stage Sampling A combination of several methods Example: What are HK college students’ favorite countries for exchange study? Cluster → Stratified Stratified → Cluster Stratified → Stratified Cluster → Cluster HKUST, MARKETING RESEARCH  Which Sampling Method to Use Method Cost/East Statistical Efficiency Convenience Very low Very poor Judgment Slight more Very poor Quota Slightly more Very poor Simple Random moderate fair Stratified costly Can be very good Cluster Less than stratified Fair to good Systematic Easier than SR Fair HKUST, MARKETING RESEARCH  SIZE AND ERROR HKUST, MARKETING RESEARCH  Error Sources in Data Collection Sampling errors Decrease with sample size We can estimate in probability samples If HKUST wants to survey student satisfaction with restaurant facilities on campus but only uses email addresses from a list that excludes part-time students, the survey results will not reflect the opinions of Non-sampling errors those part-time students. This exclusion is a non-coverage error. Non-coverage Non-response error no response Response error wrong answer Recording error write down wrong Office error HKUST, MARKETING RESEARCH  Sample Size Common question: how many respondents do we need? Historical evidence (common practice in a company) Causal research (lab experiment): at least 20-30 for each condition 10 - 15 times the number of variables in your analyses Balance between cost and benefit Research takes time and money Generally, larger sample increases accuracy (less sampling error) But the marginal benefit decreases as sample size becomes larger Does not depend on the size of the population, but the desired precision, confidence, and variation of the characteristics in the population! HKUST, MARKETING RESEARCH  Sample Size and Sampling Error How many time of a week dose a HKUST student purchase from the vending machine on campus? HKUST, MARKETING RESEARCH  not tested Population Sample Distribution of the population Draw a random sample of size n (unknown to the researcher) {X1, X2, …, Xn} Mean = 8; Std. Dev. = 2 Statistics: Sample average How good is this estimator? HKUST, MARKETING RESEARCH  Population Sample Statistics from different samples: n = 10: average = 5.4 n = 25: average = 6.2 n = 50: average = 7.8 n = 100: average = 8.1 Sampling error = difference between μ sample statistic and population parameter Sampling error decreases as n increases! Why? HKUST, MARKETING RESEARCH  Draw Random Samples of Size n Many Many Times {X1, X2, …, Xn} Sn The distribution of Sn {X1, X2, …, Xn} Sn smaller sem (sd) {X1, X2, …, Xn} Sn larger sem (sd) {X1, X2, …, Xn} Sn n=10 n=50...... “Standard Error of the Mean (SEM)” Std. Dev. of Sn = formula HKUST, MARKETING RESEARCH  How SEM Decreases as Sample Size Increases in Vending Machine Example 2.5 2 1.5 SEM 𝜎 2 1 𝑆𝐸𝑀 = = 𝑛 𝑛 0.5 0 1 11 21 31 41 51 61 71 81 91 Sample Size n HKUST, MARKETING RESEARCH  Class Exercise HKUST would like to know how satisfied their students are with restaurant facilities on campus. To measure satisfaction, they developed a questionnaire. A colleague of the marketing researcher proposes a non- probability or probability sample. HKUST, MARKETING RESEARCH  Class Exercise The marketing researcher of HKUST decides to email the questionnaire to 1,300 students. She selects the email addresses using one of the following methods from the complete list of 13,000 e-mail addresses. Which probability sampling technique does the marketing researcher propose? a. Using a computer program she randomly selects 1,300 email addresses from the complete list. simple random b. She draws one random number n between 0 and 11. Using the list of 13,000 addresses she selects the following students: n, n + 10, n + 2*10, n + 3*10, … , n + k*10, … , n + 1,299*10. systematic c. She divides the list of students in 8 different groups: business, humanities, engineering and science, and for each discipline graduate and post graduate students. From each group, she randomly selects 163 students using a computer program. Stratified d. Using the 8 groups from method c. She randomly selects 4 groups, and from these 4 groups randomly selects 325 students. cluster HKUST, MARKETING RESEARCH  Class Exercise The researcher reasons that in a probability sample many students that are selected don’t use the restaurant facilities. Which non- probability sampling techniques does the colleague propose? a. She proposes to go to one of the restaurants at campus and interview 1300 students that just had lunch. Convenience b. She proposes to interview only students that live on campus and have at least 5 dinners a week. Fortunately, she knows some students have dinner at campus every day. She suggests to interview those students first, and to ask these students to suggest other candidates that have dinner at least 5 times a week. Judgement c. Because 2/3 of all students at HKUST are males, she suggests to interview 433 females and 867 males that just had lunch at campus. Quota HKUST, MARKETING RESEARCH  Summary 1. Definitions of population and sample 2. Sampling procedures 3. Sample size and sampling error HKUST, MARKETING RESEARCH  Checklist 1. Read “Note on basic statistics” 2. Next class: Data Analysis I: Preliminary Steps Readings: chapters 16 Will hold at least 20-minute in-class project consultation 3. Project proposal due on Oct. 4th HKUST, MARKETING RESEARCH