Question 1

Which type of validity can be affected by missing values problems?

Accepted Answer

Missing values can affect all three types of validity: statistical conclusion validity (by affecting the statistical tests), internal validity (by introducing biases that can affect the causal relationship between variables), and external validity (by limiting the generalizability of the findings if the missing data is not random).

Question 2

Which of the following statements is true?

Accepted Answer

The pattern of missingness (such as whether it is missing completely at random, missing at random, or non-ignorable) can affect the conclusions drawn from the data, and thus it is often more important to address than the amount of missing values. Option B is incorrect because missing value strategies can be used for both quantitative and categorical variables. Option C is incorrect because missing values can affect both Type I and Type II errors. Option D is incorrect because missing values can be addressed for both independent and dependent variables.

Question 3

If people who were obese were more likely than others to skip a question about their weight, the missing data would be:

Accepted Answer

If the missing data is related to the characteristic being measured (in this case, weight), then it is considered missing not at random. If obese individuals were more likely to skip the question about their weight, this would fall under the category of missing not at random.

Question 4

Suppose that people who were obese were just as likely as others to skip a question about their weight, but that women were more likely than men to omit this information. In this situation, the missing data would be:

Accepted Answer

The missing data is not completely at random because it is related to gender. However, it is missing at random because obesity status is not related to the missingness. Therefore, the missing data is missing at random.

Question 5

Which pattern of missingness yields unbiased parameter estimates, regardless of which method of addressing the problem is used?

Accepted Answer

Missing completely at random (MCAR) is the only pattern of missingness that yields unbiased parameter estimates, regardless of which method is used to address the problem. This is because in MCAR, the probability of missingness is unrelated to both the observed and unobserved data. Therefore, any biases that may arise from missing data are random and not related to any underlying characteristics of the data.

Question 6

Which approach to solving missing values problems does not reduce statistical power?

Accepted Answer

Imputation methods replace missing values with estimated values based on known values, allowing for a complete dataset to be analyzed without reducing statistical power. Deletion methods, on the other hand, remove observations with missing values and therefore reduce the sample size and potentially the power of the analysis. The choice among specific imputation methods will depend on the particular dataset and research question.

Question 7

Which approach is called complete case analysis?

Accepted Answer

Complete case analysis, also known as listwise deletion, is the approach where all cases with missing data are excluded from the analysis. This approach may lead to biased results and loss of statistical power, but it is the best choice when the missingness is completely at random (MCAR) and the sample size is large enough to still have a sufficient sample size after the exclusion.

Question 8

Which approach is a reasonable approach to imputing item values on a multi-item scale?

Accepted Answer

The answer of Which approach is a reasonable approach to...

Question 9

Which imputation approach is most often used to estimate outcomes in a randomized clinical trial?

Accepted Answer

The most common imputation approach used in randomized clinical trials is Last Observation Carried Forward (LOCF) because it is simple and straightforward. LOCF imputes missing values by carrying forward the last observed value of the outcome variable for each participant. However, LOCF has been criticized for potentially inflating treatment effects and underestimating the standard error of estimates. Other imputation approaches such as regression imputation, multiple imputation, or mixed models may provide more accurate estimates, but they are more complex and require additional assumptions. Mean substitution or worst-case imputation are not recommended as they may introduce bias and distort the true relationship between the outcome and the covariates.

Question 10

When researchers in a clinical trial analyze outcomes for all study participants who were randomly assigned to a treatment group, regardless of whether they received the treatment or dropped out of the study, the analysis is called:

Accepted Answer

An intention to treat analysis includes all randomized participants, even those who did not receive the treatment or dropped out of the study. This approach is important because it preserves the randomization and avoids selection bias. A complete case analysis only includes participants who completed the study, which can introduce bias if those who dropped out are systematically different from those who remained. An available case analysis only includes participants who have complete data, which can also introduce bias.

Question 11

Which approach would replace missing values based on using an equation such as the following: Missing Value = 10.0 + (2.5 * VAR1) + (1.8 * VAR2) + (3.0 * VAR3)?

Accepted Answer

The approach described is regression imputation, which involves using a regression equation to estimate missing values based on the relationships between the missing variable and other variables in the dataset. This approach is appropriate when there is a strong relationship between the missing variable and other variables, as indicated by the high regression coefficients in the equation. Mean substitution and conditional mean substitution are simpler methods that may be appropriate when there are only a few missing values or when the missing values are not strongly related to other variables in the dataset. Regression imputation with error is a more complex approach that accounts for the variability in the imputed values, but it may not be necessary in cases where the missing values are relatively few and the imputation equation is well-specified.

Question 12

Which of the following is not a type of single imputation?

Accepted Answer

All three options listed (regression imputation, stochastic regression imputation, and expectation-maximization imputation) are types of single imputation. Therefore, D is the correct answer.

Question 13

In multiple imputation, m represents:

Accepted Answer

The letter "m" in multiple imputation refers to the number of datasets that are created. Typically, multiple datasets are created, each with slightly different imputed values, and the results are combined to produce a final imputed dataset. This helps to account for the uncertainty in the imputed values and produces more accurate estimates.

Question 14

If VAR1 has 5% missing, VAR2 has 5% missing, and VAR3 has 10% missing, which of the following is true?

Accepted Answer

The percentage of participants with no missing values cannot be directly calculated from the individual missing percentages of VAR1, VAR2, and VAR3 because the missing values could overlap or be completely separate among the variables. Therefore, the actual percentage of participants with no missing values could range between 80% and 90%, assuming the worst-case scenario where no missing values overlap, to the best-case scenario where all missing values overlap.

Question 15

Which of the following strategies is advisable in addressing missing values problems?

Accepted Answer

Sensitivity analyses help assess the impact of different missing data strategies, ensuring robustness of results.

Quiz 14: Missing Values