W2 Flashcards

Question

What is the formula for observed score?

Answer 1

X₀ = Xₜ + Xₛ + Xᵣ (Observed score = True score + Systematic error + Random error)

Answer 2

Because participants might misunderstand a question, get distracted, or have a bad day — so we can't assume perfect accuracy.

Answer 3

Consistent errors that are off-target but grouped together — like everyone slightly misinterpreting a question the same way.

Answer 4

Errors scattered all over the place with no clear pattern — unpredictable and inconsistent mistakes.

Answer 5

It checks if items that are supposed to measure the same thing actually correlate highly with each other.

Answer 6

High: ≥ 0.6, Moderate: 0.2–0.6, Low: < 0.2

Answer 7

We want items from the same construct (like B001, B002, B003 for 'willingness to pay') to have high correlations.

Answer 8

It’s a problem for convergent validity — it means the items may not be measuring the same thing.

Answer 9

It checks that items from different constructs (e.g., 'willingness to pay' vs. 'environmental concern') have low correlation.

Answer 10

That the items are unrelated as they should be — this supports the discriminant validity of the constructs.

Answer 11

Low correlations (e.g., under 0.2 or even negative, like -0.2 to -0.8) to show that the constructs are distinct.

Answer 12

Reliability is about consistency and has little random error. Validity is about measuring the right thing. Reliability is required for validity.

Answer 13

It checks if multiple items measuring the same concept give consistent results.

Answer 14

You split the items in a scale into two halves and see if the results from both halves are similar.

Answer 15

A measure of how closely related a set of items are. It uses the number of items and their average correlation.

Answer 16

It means items are strongly related, showing high internal consistency.

Answer 17

0 to 0.6 = bad, - 0.6 to 0.7 = moderate, - 0.7 to 0.8 = sufficient, - 0.8 to 0.9 = very good, - 0.9 to 1 = suspiciously high

Answer 18

It measures the level of agreement across test items — higher values mean better reliability.

Answer 19

α = (N * 𝑐̄) / (v̄ + (N − 1) * 𝑐̄), where N = number of items, 𝑐̄ = average covariance, and v̄ = variance

Answer 20

Use the function alpha() — it is commonly used when doing factor analysis.

Answer 21

It helps simplify many variables into a few factors that represent patterns in the data.

Answer 22

1. Reduce many variables into fewer factors 2. Structure correlating variables 3. Understand unmeasured constructs 4. Handle measurement errors

Answer 23

The entire group of people, items, or events you want to make conclusions about.

Answer 24

A complete measurement of every element in the population.

Answer 25

A subgroup of the population that is selected to represent the whole group.

Answer 26

Define the target population — the full group that has the info the researcher is looking for.

Answer 27

A single member of the population, like one person or one household.

Answer 28

The specific item or person available for selection, such as a household or a phone number.

Answer 29

Determine the sampling frame — a list or method that helps you identify and reach the target population.

Answer 30

Customer databases, phonebooks, or lists of registered users.

Answer 31

It helps you reach your target population accurately, but it can be hard to get or outdated.

Answer 32

Missing people like those without phone numbers or those who bought something anonymously can skew the results.

Answer 33

You can choose a non-probability sampling method, redefine your population, screen participants, or adjust your data.

Answer 34

Once your sampling frame is fully adequate — if not, fix it or switch to non-probability methods.

Answer 35

Sampling methods where each element has a known chance of being selected. Results can be generalized to the population.

Answer 36

Methods where elements do not have a known chance of being selected. Less generalizable but easier and cheaper.

Answer 37

You pick the easiest people to reach, like students or online surveys. It’s quick and cheap, but not representative.

Answer 38

You choose participants based on expert judgment. Useful but subjective and not very generalizable.

Answer 39

You select people to match a set quota (e.g., 50% male, 50% female). Easy to control groups but not fully representative.

Answer 40

You start with a few participants and ask them to refer others. Great for rare populations but highly biased.

Answer 41

Everyone has an equal chance to be selected. It’s fair and easy to understand but doesn’t guarantee perfect representation.

Answer 42

You choose every nth person from a list (e.g., every 10th). It’s simple but can be biased if there’s a pattern in the list.

Answer 43

You divide the population into subgroups (strata) and sample from each. It’s precise but takes time.

Answer 44

The sample size from each subgroup matches its proportion in the population.

Answer 45

You sample more or less from some subgroups than their population size suggests, based on research needs.

Answer 46

You split the population into clusters and randomly select entire clusters. It’s cheap and easy but less accurate.

Answer 47

You first pick clusters, then randomly select individuals or households within those clusters. Useful for large areas.

Answer 48

Example: First pick random city blocks, then pick every 5th or 7th house in each block.

Answer 49

It happens when the sampling frame doesn’t match the target population. Some people are left out or wrongly included.

Answer 50

(a) Under-coverage: true population members are excluded (b) Miss-coverage: non-population members are included

Answer 51

You can sometimes ignore it, but with large samples you should redefine the population based on the frame.

Answer 52

It’s a deliberate deviation caused by the sampling technique — often due to practical limits like cost or access.

Answer 53

(a) Interest in specific groups (e.g., quota sampling) (b) Lack of access to some people (e.g., snowball sampling) (c) Budget issues (e.g., cluster sampling)

Answer 54

Because both coverage and sampling errors can mess up your results if not handled properly.

Answer 55

Confidence and power levels (higher levels need bigger samples) - Effect size (smaller effects need bigger samples) - Survey complexity (complex surveys need bigger samples) - Model complexity (complex models need bigger samples)

W2 Flashcards

(79 cards)