- A researcher studying medication adherence finds that 15% of patients fail to complete follow-up questionnaires. Analysis reveals that missingness is associated with patient age and baseline health status (both fully observed), but shows no association with the actual adherence rates when controlling for these factors. However, the researcher suspects that patients with poor adherence may be less likely to respond regardless of their age or health status. To definitively classify this missing data mechanism, the researcher would need to:
Answer: Obtain information about adherence rates among the non-responders through alternative data sources
To definitively distinguish between MAR and MNAR, the researcher needs data on the missing values themselves. The current analysis suggests MAR (missingness explained by observed variables), but the suspicion of MNAR (poor adherence causing non-response) can only be confirmed by obtaining the missing adherence data through medical records, pharmacy data, or other sources. Option A describes sensitivity analysis, which is useful but doesn't definitively classify the mechanism. Option B has already been done. Option D addresses power but not missing data classification.
- A researcher collects data on both self-reported and objectively measured physical activity. Self-reported data are missing more frequently among participants with low objective measurements, but objective data are missing completely at random due to device failures. When analyzing the relationship between self-reported and objective measures, what is the most important consideration regarding missing data?
Answer: The MAR mechanism for self-reported data allows for valid imputation using objective measures
When analyzing missing data patterns, you need to identify the underlying mechanisms and choose strategies that leverage available information while minimizing bias.
In this scenario, you have two distinct missing data mechanisms. The objective measurements are missing completely at random (MCAR) due to random device failures - the missingness has no relationship to any observed or unobserved variables. The self-reported data follows a missing at random (MAR) pattern - it's systematically missing based on an observed variable (low objective activity levels), but not based on the unobserved self-reported values themselves.
The key insight is that MAR data can be validly imputed using observed variables that predict the missingness pattern. Since participants with low objective measurements are more likely to have missing self-reported data, you can use the objective measurements as predictors in your imputation model. This leverages the systematic relationship to produce unbiased estimates.
Option A is incorrect because having some MCAR data doesn't justify complete-case analysis when you also have MAR data - you'd lose valuable information and introduce bias. Option C misses the point that mixed mechanisms can often be handled with a unified approach that accounts for both patterns. Option D is wrong because this isn't missing not at random (MNAR) - the missingness depends on observed objective measures, not unobserved self-reported values.
Remember: MAR data is your friend in imputation because you can model the missingness using observed predictors. Always look for variables that explain why data might be missing.
- In a study examining both anxiety and depression symptoms, researchers find that participants with missing anxiety data tend to have higher observed depression scores, while participants with missing depression data tend to have higher observed anxiety scores. Both patterns persist after controlling for demographic variables. What missing data mechanism does this cross-symptom pattern most likely represent?
Answer: MAR because missingness in each variable can be predicted from the other observed variable
When analyzing missing data patterns in biostatistics, you need to understand the three fundamental mechanisms: MCAR (Missing Completely at Random), MAR (Missing at Random), and MNAR (Missing Not at Random). The key distinction lies in whether missingness depends on observed data, unobserved data, or neither.
In this scenario, the cross-symptom pattern reveals a classic MAR mechanism. People missing anxiety data have higher observed depression scores, and people missing depression data have higher observed anxiety scores. This means the missingness in each variable is systematically related to the observed values of the other variable. Since you can predict who will have missing anxiety data based on their observed depression scores (and vice versa), this missingness is "random" conditional on the observed information—the definition of MAR.
Option A is incorrect because MCAR would show no systematic relationships between missingness and any observed variables. The clear patterns described rule this out. Option C misinterprets the mechanism—if people were avoiding reporting their more severe symptoms, you'd expect those with missing data to have higher scores in that same domain when observed, not the opposite pattern described. Option D incorrectly assumes different mechanisms are operating when the symmetric cross-pattern suggests a single MAR mechanism affecting both variables.
Study tip: For missing data questions, always ask: "Can I predict who has missing data based on what I can observe?" If yes, it's likely MAR. The cross-variable prediction pattern here is a telltale MAR signature.
- A longitudinal study of weight loss interventions finds that participants who are not losing weight as expected are more likely to skip weigh-in appointments. The study implements a protocol where research staff call participants who miss appointments to obtain self-reported weights over the phone. Analysis shows that self-reported weights from phone calls are systematically 3-5 pounds lower than clinic-measured weights for the same individuals when both are available. After implementing this protocol, the missing data mechanism for weight measurements is best described as:
Answer: MNAR because the phone-collected data introduces systematic bias related to unmeasured true weights
The mechanism remains MNAR because the intervention doesn't eliminate the fundamental problem - it just substitutes biased measurements for missing ones. The systematic underreporting in phone weights means the 'recovered' data still depends on unmeasured true weight values. Participants who skip appointments due to poor progress now provide systematically biased data rather than missing data. Option A focuses on predictability rather than the bias mechanism. Option C incorrectly suggests data are available for everyone (they're not - they're biased substitutes). Option D incorrectly assumes the intervention eliminates the underlying MNAR mechanism.
- In a longitudinal study of depression, researchers notice that participants with more severe depression scores at baseline are increasingly likely to drop out of subsequent follow-up visits. However, among participants with similar baseline depression scores, the probability of dropout appears unrelated to their unobserved future depression trajectory. What missing data mechanism does this scenario most likely represent?
Answer: Missing at Random (MAR) because dropout depends on observed baseline depression scores
Missing data mechanisms are fundamental to understanding how dropout patterns affect study validity. When analyzing longitudinal studies, you need to classify why data is missing to choose appropriate statistical methods.
This scenario describes Missing at Random (MAR) because dropout depends on observed baseline depression scores, but among participants with similar baseline scores, dropout is unrelated to their future (unobserved) depression trajectory. The key insight is that once you condition on the observed baseline data, the missingness mechanism becomes random with respect to the unobserved outcomes.
Answer B correctly identifies MAR. The dropout probability can be fully explained by the observed baseline depression scores, making the missing data mechanism "ignorable" for analysis purposes once baseline severity is accounted for.
Answer A incorrectly suggests MCAR. While dropout is unrelated to future depression trajectories within baseline severity groups, it's not completely random since it varies by baseline depression scores. MCAR would require dropout to be independent of both observed and unobserved variables.
Answer C misidentifies this as MNAR. MNAR occurs when missingness depends on the unobserved values themselves (like participants dropping out specifically because their current depression worsened). Here, dropout depends only on observed baseline scores, not unobserved future trajectories.
Answer D is incorrect because the correlation between baseline and follow-up scores doesn't determine the missing data mechanism. The mechanism depends on what drives the dropout decision, not the strength of variable relationships.
Remember: MAR means missingness can be explained by observed data alone. If you can predict who drops out using only baseline characteristics, you're likely dealing with MAR.
- In a study measuring cognitive decline in elderly patients, researchers notice that patients with more severe cognitive impairment are increasingly likely to miss follow-up assessments. The research team implements a strategy where they contact family members to obtain proxy ratings for patients who miss appointments. After implementing this strategy, the missing data pattern for cognitive scores becomes independent of all observed patient characteristics. The missing data mechanism has changed from:
Answer: MNAR to MCAR because the intervention eliminated systematic patterns in missingness
Initially, missingness depended on the severity of cognitive impairment (the outcome being measured), making it MNAR. After implementing proxy ratings, missing data became independent of all observed characteristics, indicating MCAR. The intervention effectively eliminated the systematic relationship between cognitive status and missingness by providing alternative data sources for impaired patients. Option B is incorrect because the initial mechanism was MNAR (not MAR) since missingness depended on cognitive impairment itself. Option C would be correct if missingness still showed some systematic patterns. Option D incorrectly suggests the change went from MCAR to MAR.
- A researcher tests whether data are Missing Completely at Random (MCAR) using Little's MCAR test and obtains a p-value of 0.03. The researcher concludes that the data are Missing at Random (MAR) rather than MCAR. What is the primary limitation of this reasoning?
Answer: Rejecting MCAR does not automatically confirm MAR since the data could be MNAR
Missing data analysis requires understanding three distinct mechanisms: Missing Completely at Random (MCAR), Missing at Random (MAR), and Missing Not at Random (MNAR). When you encounter Little's MCAR test questions, remember that statistical tests can only tell you what to reject, not what to accept.
Little's MCAR test specifically tests the null hypothesis that data are MCAR. With a p-value of 0.03 (< 0.05), you reject the MCAR hypothesis, meaning the missingness pattern is not completely random. However, rejecting MCAR leaves two possibilities: the data could be MAR (missing at random, but dependent on observed variables) or MNAR (missing not at random, dependent on unobserved values). The test cannot distinguish between these two alternatives, so concluding the data are MAR is premature and illogical.
Option A is wrong because p = 0.03 is statistically significant at the conventional α = 0.05 level, providing sufficient evidence to reject MCAR. Option B mischaracterizes the test's purpose—Little's MCAR test isn't designed to distinguish MAR from MNAR; it only tests whether data deviate from MCAR. Option D incorrectly suggests a power issue when the significant result actually indicates adequate power to detect the deviation from MCAR.
Study tip: Remember that statistical tests tell you what to reject, not what to conclude positively. When you reject one hypothesis in a multiple-alternative scenario, you haven't automatically confirmed any specific alternative. Always consider what remains possible after rejection.
- A researcher studying academic performance finds that students with lower GPA are more likely to have missing data for self-reported study hours. The researcher argues this is MAR because GPA is observed. However, a colleague suggests it might be MNAR because students who actually study very little may be more likely to skip the question regardless of their GPA. What additional information would best help distinguish between these interpretations?
Answer: Whether students with similar GPAs but different actual study hours show different missingness rates
When you encounter questions about missing data mechanisms, you need to distinguish between MAR (Missing At Random) and MNAR (Missing Not At Random) by understanding what drives the missingness pattern.
The key insight is that MAR means missingness depends only on observed variables, while MNAR means missingness depends on the unobserved values themselves. In this scenario, the researcher claims it's MAR because missingness relates to observed GPA. The colleague suggests it's MNAR because students who actually study little may skip the question regardless of GPA.
Answer D is correct because it directly tests the MNAR hypothesis. If you compare students with similar GPAs but examine whether their actual study hours affect their likelihood of missing data, you can determine if the unobserved variable (actual study time) drives missingness. If students with identical GPAs show different missingness rates based on their true study hours, this confirms MNAR.
Answer A is wrong because correlations in complete cases don't reveal what drives missingness in the incomplete cases. Answer B is incorrect because controlling for other academic variables still wouldn't distinguish whether actual study hours (the missing variable itself) affects missingness. Answer C is wrong because the overall proportion of missing data tells you nothing about the mechanism causing the missingness.
Remember this pattern: To distinguish MAR from MNAR, look for tests that examine whether the unobserved values themselves influence the probability of being missing, holding observed covariates constant.
- A survey about workplace harassment includes a question about whether the respondent has personally experienced harassment. This question has a 25% non-response rate, while other sensitive questions (salary, personal relationships) have only 5% non-response rates. What does this pattern most strongly suggest about the missing data mechanism for the harassment question?
Answer: MNAR because individuals who experienced harassment may be more likely to skip this specific question
When you encounter questions about missing data patterns, focus on what drives the non-response behavior. Missing data mechanisms fall into three categories: MCAR (missing completely at random), MAR (missing at random), and MNAR (missing not at random). The key is understanding whether the "missingness" itself relates to the unobserved values.
The dramatically higher non-response rate for the harassment question (25% vs. 5%) suggests that people's decision to skip this question is directly related to their harassment experience. Those who experienced harassment are likely more hesitant to answer due to trauma, fear of identification, or emotional difficulty, while those who haven't experienced harassment may feel comfortable answering. This creates a systematic relationship between the missing data and the unobserved values themselves—the hallmark of MNAR.
Option A is wrong because MCAR requires missingness to be completely random, unrelated to any variable. The pattern here is clearly systematic, not random. Option B incorrectly suggests MAR, which would mean non-response could be predicted from other observed variables (like demographics), but the evidence points to the harassment experience itself driving non-response. Option D misses the point—while demographic predictors might help confirm the mechanism, the stark difference in response rates already provides strong evidence of the underlying mechanism.
Remember: MNAR occurs when the probability of missing data depends on the unobserved values themselves. In surveys about sensitive personal experiences, be especially alert for MNAR patterns where those most affected by the topic are most likely to avoid responding.
- In a clinical trial comparing two treatments, blood pressure measurements are missing for some patients. The missing data analysis shows: (1) Missingness is independent of treatment group, age, and baseline BP; (2) Patients with missing data have similar demographic characteristics to those with complete data; (3) The proportion of missing data is constant across all measurement timepoints. Despite these findings, complete case analysis and multiple imputation yield substantially different treatment effect estimates. This discrepancy most likely indicates:
Answer: The missing data mechanism is MNAR and the assumption underlying multiple imputation has been violated
Despite evidence suggesting MCAR (points 1-3), the substantial difference between complete case analysis and multiple imputation results suggests MNAR. Under true MCAR, both methods should yield similar estimates (complete case gives unbiased estimates under MCAR, just with less precision). The discrepancy indicates that missingness likely depends on unobserved blood pressure values themselves. Multiple imputation assumes MAR, so if data are MNAR, imputation will be biased. Options A and D incorrectly assume MCAR would cause this discrepancy. Option B incorrectly suggests the issue is with imputation model specification rather than the missing data mechanism.
- A longitudinal study tracks both self-reported pain levels and objective biomarkers of inflammation. At each time point, participants with missing self-reported pain data show significantly higher inflammation biomarkers than those with complete pain data. However, participants with missing biomarker data show inflammation levels that are indistinguishable from the overall sample when measured at adjacent time points. How should the missing data mechanisms be characterized?
Answer: MAR for pain data and MCAR for biomarker data based on different predictability patterns
When analyzing missing data in longitudinal studies, you need to distinguish between three mechanisms: MCAR (Missing Completely At Random), MAR (Missing At Random), and MNAR (Missing Not At Random). The key is examining whether missingness patterns relate to observed or unobserved variables.
For the pain data, participants with missing responses show significantly higher inflammation biomarkers. This creates a clear pattern where missingness is predictable from observed inflammation levels, indicating MAR. The missing pain reports aren't random—they're systematically related to high inflammation, which you can observe and account for in your analysis.
For the biomarker data, participants with missing values show inflammation levels indistinguishable from the overall sample at adjacent time points. This suggests no systematic pattern—the missingness appears unrelated to any observable characteristics, indicating MCAR.
Option A incorrectly assumes both variables follow the same mechanism simply because each could theoretically be predicted. Option C suggests MNAR for pain data, but MNAR would require the missingness to depend on the unobserved pain values themselves, not the observed inflammation markers. Option D incorrectly assumes longitudinal correlation automatically makes missing data MCAR across both variables.
The correct characterization is B: MAR for pain data (predictable from observed inflammation) and MCAR for biomarker data (no observable pattern).
Study tip: Always examine what observed variables relate to missingness patterns. If you can predict missingness from observed data, it's MAR. If there's no detectable pattern, consider MCAR. Reserve MNAR for when missingness depends on the unobserved values themselves.
- During the trial, 18% of participants in the treatment group and 12% of participants in the control group discontinue before the 12-week assessment. Exit interviews reveal that treatment group discontinuations are primarily due to side effects (nausea, headaches), while control group discontinuations are mainly due to lack of improvement. In both groups, participants who discontinue have similar baseline characteristics to those who complete the study. The missing 12-week HAM-D data in this scenario represents:
Answer: MNAR because discontinuation decisions are related to treatment response, which is directly related to the outcome being measured
When analyzing missing data patterns in clinical trials, you need to determine whether the missingness mechanism depends on unobserved outcomes. The key distinction is whether the reason for dropout is related to what the missing outcome value would have been.
In this scenario, the treatment group participants are discontinuing primarily due to side effects, while control group participants are dropping out due to lack of improvement. Both of these reasons are directly tied to treatment response, which correlates with the HAM-D depression scores being measured. Participants experiencing side effects might have been improving but couldn't tolerate the medication, while those lacking improvement would likely have higher (worse) depression scores. Since the probability of missingness depends on the unobserved HAM-D values themselves, this represents Missing Not at Random (MNAR).
Option B incorrectly assumes that documented reasons and balanced baseline characteristics make data MAR. However, MAR requires that missingness depends only on observed data, not on the unobserved outcomes. Option C misinterprets MCAR, which requires missingness to be completely random and unrelated to any variables. The fact that dropout reasons differ systematically between groups rules this out. Option D confuses the ability to model different dropout rates with the underlying missingness mechanism - even if you can account for different rates, the data remains MNAR when dropout reasons relate to unobserved outcomes.
Study tip: Focus on whether dropout reasons connect to what the missing outcome values would have been, not just whether you can document or model the dropout patterns.
- A researcher studying workplace stress finds that employees with higher stress levels are more likely to have missing data for income questions on a survey. Additionally, some employees randomly failed to receive the income section due to a printing error that affected every 10th survey booklet. What type of missing data mechanism best describes the missing income data in this study?
Answer: A combination of MCAR and MAR mechanisms operating simultaneously
When you encounter questions about missing data mechanisms, you need to identify what causes the missingness and whether that cause is observed, unobserved, or completely random.
This scenario presents two distinct mechanisms operating simultaneously. The printing error affects every 10th survey booklet regardless of any participant characteristics—this represents Missing Completely at Random (MCAR) because the missingness is purely due to chance and unrelated to any variables in the study. Meanwhile, employees with higher stress levels are more likely to skip income questions, which represents Missing at Random (MAR) because missingness depends on an observed variable (stress level) but not on the missing variable itself (income).
Answer A is incomplete because it only acknowledges the printing error component while ignoring the stress-related missingness pattern. Answer B similarly focuses on just one mechanism, recognizing the MAR component but overlooking the MCAR printing error. Answer C mischaracterizes the situation as Missing Not at Random (MNAR), which would require income levels themselves to directly influence whether income data is missing—but the scenario indicates stress levels, not income amounts, drive the non-response pattern.
Answer D correctly identifies that both MCAR and MAR mechanisms are operating together. Real-world studies frequently involve multiple missing data mechanisms simultaneously, and you must recognize all patterns present rather than forcing the situation into a single category.
Remember: missing data questions often involve multiple mechanisms. Read carefully to identify all sources of missingness and classify each one separately before determining the overall pattern.
- In a longitudinal study examining factors associated with depression scores, participants with higher baseline anxiety levels are more likely to drop out before the final assessment. Additionally, participants who develop severe depression during the study are also more likely to withdraw. The probability of missingness depends on both the observed anxiety scores and the unobserved depression severity at the time of dropout. This missing data pattern is best classified as:
Answer: Missing Not at Random (MNAR) because missingness depends on the unobserved depression severity
This is MNAR because the missingness depends on the unobserved depression severity at dropout. While anxiety is observed and associated with dropout, the key factor is that participants with severe depression (unobserved at dropout) are more likely to withdraw. When missingness depends on the unobserved values of the variable being studied, it's MNAR. Option A is wrong because dropout isn't random. Option B is incomplete - while anxiety predicts dropout, the unobserved depression severity is the crucial factor. Option D misunderstands that study design alone doesn't determine MAR classification.
- A nutritionist studying eating habits finds that participants with eating disorders are more likely to have missing data for questions about caloric intake, but not for questions about exercise habits. Among participants without eating disorders, missing data patterns appear random for both types of questions. How should this missing data pattern be characterized?
Answer: MAR overall because eating disorder status is observed and can predict missingness
When analyzing missing data patterns, you need to consider three key mechanisms: MCAR (Missing Completely At Random), MAR (Missing At Random), and MNAR (Missing Not At Random). The critical distinction lies in whether you can predict missingness from observed variables.
In this scenario, eating disorder status is an observed variable that predicts missing caloric intake data. Since you can identify which participants are more likely to have missing data based on this observable characteristic, the missingness follows a predictable pattern that can be accounted for in your analysis. This defines MAR - the missingness depends on observed covariates but not on the unobserved values themselves.
Answer A incorrectly splits the mechanisms by question type rather than recognizing that the overall pattern across all participants is MAR. While exercise data might appear MCAR within each group, the presence of eating disorder status as a predictor makes the overall mechanism MAR.
Answer B suggests MNAR, which would apply if the missing caloric intake data depended on the actual caloric values themselves (e.g., people with extremely high intake refusing to report). However, since eating disorder status is observed and explains the missingness, this is MAR, not MNAR.
Answer D overcomplicates the classification. While the description might seem accurate, the standard approach is to characterize the overall mechanism as MAR when an observed variable predicts missingness patterns.
Remember: if you can predict who has missing data using variables you've observed, it's MAR. This distinction is crucial because MAR allows for valid statistical correction methods.
- In a study measuring both income and job satisfaction, income data are missing more frequently for participants with low job satisfaction scores. However, when the analysis is stratified by education level, this relationship between job satisfaction and missing income data disappears within each education stratum. What does this pattern suggest about the missing data mechanism?
Answer: The data are MAR with education as a key covariate for modeling missingness
When analyzing missing data patterns, you need to understand the three key mechanisms: MCAR (Missing Completely at Random), MAR (Missing at Random), and MNAR (Missing Not at Random). The critical distinction lies in whether missingness depends on observed or unobserved variables.
This scenario demonstrates a classic MAR pattern. Initially, income data appears to be missing more often for participants with low job satisfaction, suggesting the missingness isn't random. However, when you stratify by education level, this relationship disappears within each education stratum. This means education is acting as a confounding variable that explains the apparent relationship between job satisfaction and missingness. Once you account for education, the missingness becomes random within each education group.
Option A is incorrect because the data aren't MCAR overall – there's still a systematic pattern of missingness that depends on observed variables (education). MCAR would mean no relationship exists at any level of analysis. Option C misinterprets the situation as MNAR, but since the relationship between job satisfaction and missingness disappears when conditioning on an observed variable (education), this isn't MNAR. MNAR would require missingness to depend on the unobserved income values themselves. Option D is wrong because we can determine the mechanism – the pattern clearly indicates MAR once we identify education as the key variable.
Remember: When missingness patterns change after stratifying by observed covariates, you're likely dealing with MAR. The key is identifying which observed variables explain the missingness pattern.
- In a longitudinal study, researchers observe that participants who miss the 6-month follow-up visit are more likely to also miss the 12-month visit, even after controlling for all baseline characteristics. Among participants who attended the 6-month visit, attendance at 12-month follow-up appears random. What missing data mechanism is most likely operating for the 12-month visit data?
Answer: MAR because 6-month attendance status is observed and predicts 12-month attendance
When you encounter missing data questions in biostatistics, focus on what information predicts missingness and whether that information is observed or unobserved.
Missing data mechanisms depend on the relationship between missingness and other variables. Here, the key insight is that 6-month attendance status is observed and strongly predicts 12-month attendance, even after controlling for baseline characteristics. This creates a situation where missingness depends on observed data.
Answer B is correct because this scenario perfectly describes Missing at Random (MAR). Under MAR, the probability of missing data can depend on observed variables—in this case, whether someone attended the 6-month visit. Since we know each participant's 6-month attendance status, we have the information needed to account for this pattern of missingness in our analysis.
Answer A misinterprets the situation by focusing only on the subset who attended at 6 months, ignoring the overall pattern. Answer C suggests Missing Not at Random (MNAR), but MNAR requires that missingness depends on unobserved factors. Here, the predictor (6-month attendance) is fully observed. Answer D incorrectly assumes we need to know participants' reasons for missing visits, but missing data mechanisms are defined by statistical relationships, not underlying motivations.
Study tip: For missing data questions, always ask: "What predicts missingness, and is that predictor observed?" If the predictor is observed, it's MAR. If it's unobserved or the missing values themselves, it's MNAR. True randomness (MCAR) is rare in real studies.
- In a multi-site clinical trial, Site A experiences a data server failure that results in loss of Week 8 assessment data for all participants enrolled during a specific two-week period, regardless of their treatment assignment or clinical characteristics. Site B has missing Week 8 data primarily for participants who experienced adverse events and withdrew from the study. Site C has missing data for participants whose Week 8 values would have been in the extreme ranges (very high or very low) due to a data validation algorithm that flagged and excluded outliers. The missing data mechanisms at Sites A, B, and C are best classified as:
Answer: MCAR, MNAR, and MNAR respectively
Site A is MCAR because server failure affected all participants during a specific period regardless of any characteristics. Site B is MNAR because withdrawal due to adverse events likely depends on the unobserved outcome values themselves (patients with worse outcomes withdraw). Site C is MNAR because the validation algorithm excluded data based on the extreme values of the outcome variable itself. Option A incorrectly classifies Sites B and C as MAR. Options A and C fail to recognize that withdrawal due to adverse events often depends on unobserved outcomes. Option C incorrectly classifies the random server failure as MAR.
- In a multi-site clinical trial, Site A has 10% missing outcome data, Site B has 15% missing data, and Site C has 30% missing data. Investigation reveals that Site C had staffing shortages during data collection, while Sites A and B had adequate staffing. Within each site, missingness appears unrelated to patient characteristics or treatment assignment. What is the most appropriate missing data classification for the primary analysis?
Answer: MAR because site is observed and predicts missingness probability systematically
When analyzing missing data patterns in biostatistics, you need to classify the mechanism to choose appropriate analytical methods. The three main categories are MCAR (Missing Completely at Random), MAR (Missing at Random), and MNAR (Missing Not at Random).
The correct answer is B because site is an observed variable that systematically predicts missingness probability. Site C's 30% missing rate due to staffing shortages creates a clear pattern where knowing the site tells you the likelihood of missing data. Since site is observed and can be included in your analysis model, this makes the missingness MAR. The key insight is that while missingness appears random within each site, it's systematically different between sites due to the operational factor (staffing).
Option A is incorrect because MCAR requires missingness to be completely random across the entire study, not just within sites. The dramatic difference between sites (10%, 15%, vs 30%) violates this assumption.
Option C misapplies MNAR, which occurs when missingness depends on the unobserved outcome itself. Here, missingness depends on an observed variable (site) and operational factors, not on unmeasured patient outcomes.
Option D overcomplicates the situation. While different sites have different missing rates, this doesn't necessarily require different analytical approaches since site can be incorporated as a covariate in standard MAR methods.
Remember: MAR doesn't mean "random everywhere" – it means missingness is random conditional on observed variables. If an observed factor (like study site) predicts missingness patterns, classify it as MAR and include that factor in your analysis.
- In an online survey about internet usage habits, the survey software randomly crashes for 8% of participants during completion, and separately, participants who spend more time online are more likely to abandon the survey before reaching questions about problematic internet use. What is the most accurate characterization of the missing data mechanism for the problematic internet use questions?
Answer: Combined MCAR and MNAR mechanisms affecting the same outcome variable
When analyzing missing data mechanisms in surveys, you need to consider whether the probability of missingness depends on observed data, unobserved data, or neither. This question presents a scenario with two distinct causes of missing data operating simultaneously.
The correct answer is D because you have two different mechanisms creating missingness for the same outcome variable. The software crashes represent a Missing Completely At Random (MCAR) mechanism since technical failures occur independently of participant characteristics or responses. Meanwhile, participants with problematic internet use avoiding relevant questions represents a Missing Not At Random (MNAR) mechanism because the missingness depends on the unobserved values themselves—those with problematic use are systematically more likely to have missing data on that topic.
Option A is incorrect because while software crashes are indeed MCAR, this ignores the systematic avoidance by heavy internet users. Option B mischaracterizes the avoidance mechanism as Missing At Random (MAR)—even though online time might be measurable, the avoidance specifically relates to the unobserved problematic use patterns, not just general online time. Option C incorrectly dismisses the random software crashes, focusing only on the systematic avoidance.
When you encounter missing data questions, always ask: "What causes the missingness?" If multiple distinct causes operate on the same variable—especially when they follow different mechanisms—the overall pattern reflects a combination of those mechanisms. Look for scenarios where technical issues (usually MCAR) combine with systematic participant behavior (often MNAR).
