Market Research MT Slides 2

Scientific observation

recording patterns of behaviors without communicating with those involved (natural vs. contrived, disguised vs. undisguised, human vs. machine)

Natural vs contrived observation

natural: typical environment, contrived: artificial environment

Disguised vs undisguised observation

there as yourself vs. there as someone else (pose as customers, shop at competition, collect observations, etc)

Human vs mechanical observation

eye trackers, wearable tech, TV audimeters, app tracking, cookies

population of interest

people you want to generalize to (canadian consumers, etc.)

study population

people you can get access to for your study

study sample

people you actually study (we want to use the sample to make inferences about the population of interest)

Probability sampling

every element of the population has a known chance of being selected for the sample

Non-probability sampling

convenience

Simple random sampling (probability)

Randomly draw 5 people out of 30 (often not true randomness unless a random number generator used)

Systematic random sampling

entire population numbered and elements selected using a skip interval (every nth name is selected)

Stratified random sampling

Divide population into mutually exclusive groups and do simple random sample from each (old, middle, young) (select # from each group that’s reflective of the population)

Cluster random sampling

Partition population into mutually exclusive clusters and do simple random samples of the clusters (eg. maple street, 4th street, and 1st street) — sampple groups from different schools in a district and compare them) — each cluster is representative of the population as a whole

Convenience non-random sampling

Using people easily accessible, like mall intercepts or high-traffic locations, posting links on social media

Judgement non-random sampling

selection criteria based on researcher’s personal judgement about the study population (learning about extreme cases or targeting certain people)

Snowball non-random sampling

selected based on referrals form initial respondents (when sample unit characteristics are rare or hard to find)

Quota non-random sampling

based on demographic or classification factors, quotas are established for population sub-groups

Sampling and response bias

goal for no response bias, but we face 1) selection bias (no low-income voters) and 2) non-response bias (only less busy people vote)

Sampling error

The goal is to be “95% sure” that the real level of support in the whole population is captured in the estimate

95% confidence interval

if repeated across many independent samples from the same population, it would produce intervals containing the true population parameter 95% of the time

Bigger sample size =

smaller margin of error (but decreasing returns to accuracy and dramatic cost increases)

Sample size for mean

Z²o²/E² (Z=confidence interval (1=68%, 2=95%, 3=99.7%), o=standard dev, E=acceptable sampling error)

Sample size for proportion

Z²(P(1-P))/E² (P=estimate of true population proportion)

Catch-22 of sample size

How to calculate o or p before data collection? (secondary data, pilot sample, intuition)

Law of Small Numbers

small samples produce wildly variable estimates and are less likely to be representative of the population