Lecture 2: Missing Data (Alt 2) Flashcards
(35 cards)
What can happen if missing data is not properly addressed in psychological research?
It can seriously distort study results.
What example is used to demonstrate the impact of different patterns of missingness on results?
Reported lifetime sexual partners, demonstrating how selective nonresponse due to stigma can drastically alter means and create misleading conclusions.
How does the lecturer distinguish missing data from selection bias?
Missing data refers to cases where participants omit specific responses, whereas selection bias refers to systematic issues in who participates in the study.
What can selection bias lead to, according to the lecture?
Spurious correlations, such as those seen in examples involving acting talent and physical attractiveness, American college athletes, and health outcomes among smokers.
What mechanism explains how selection bias creates false correlations?
Joint selection on two variables.
What are the common sources of missing data discussed in the lecture?
Entire questionnaires being left incomplete, missing values on specific variables, and selective nonresponse from subsets of participants.
What three categories of missing data are distinguished in the lecture?
Missing Completely at Random (MCAR), Missing at Random (MAR), and Missing Not at Random (MNAR).
What is the impact of MCAR and MAR on parameter estimates and statistical power?
MCAR and MAR do not bias parameter estimates but reduce statistical power and precision.
What is the impact of MNAR on parameter estimates?
MNAR biases estimates and is difficult to detect and correct.
What example is given for MCAR?
A random programming glitch.
What example is given for MAR?
Older people not responding to sexuality questions.
What characterises MNAR, as discussed in the lecture?
The missingness is related to the unobserved value itself.
How should researchers respond to missing data issues according to the lecture?
They should not hide them but transparently report them to enhance credibility.
What is the first practical step in identifying missingness?
Descriptive checks such as per-participant and per-variable missingness rates.
What statistical test is introduced to examine the randomness of missing data?
Little’s MCAR test.
What does a significant result on Little’s MCAR test suggest?
Systematic missingness.
What follow-up analysis is suggested after a significant MCAR test result?
Separate variance t-test to determine if the probability of missing data is predicted by a specific variable.
What limitation is noted about statistical tests for missingness?
They are limited to measured variables, and large samples may yield trivial significance.
What are the two broad strategies for handling missing data discussed in the lecture?
Deletion and substitution.
What is listwise deletion?
Removing any case with missing data from all analyses — appropriate when less than 5% of data is missing.
What is pairwise deletion?
Removing cases only from specific analyses that include the missing variable.
What is a potential issue with pairwise deletion?
It can result in inconsistencies across analyses.
What is a key limitation of mean substitution?
It underestimates variability and inflates false positives.
What is a flaw in regression substitution?
It underestimates standard errors.
