Types of Missing Data and Estimation Methods

Study Notes

Types of missing data:
- MCAR (Missing Completely At Random): Missing values are unrelated to the data.
- MAR (Missing At Random): Missing values are related to observed data, but not to the missing data itself.
- NMAR (Not Missing At Random): Missing values are related to the missing data itself.
Estimation methods:
- Listwise deletion: Delete rows with missing values.
- Pairwise deletion: Delete rows with missing values for each pair of variables.
- Regression imputation: Impute missing values using a regression model.

Types of errors:
- Data entry errors: Incorrect or inaccurate data entry.
- Measurement errors: Errors in data collection or measurement.
- Data processing errors: Errors in data processing or storage.
Data cleaning techniques:
- Handling outliers: Identify and handle outliers to prevent data corruption.
- Data normalization: Normalize data to prevent differences in scales.
- Data transformation: Transform data to prevent differences in distributions.

Types of missing data extraction:
- List extraction: Extract a list of rows with missing values.
- Pair extraction: Extract pairs of variables with missing values.
Extraction methods:
- SQL queries: Use SQL queries to extract missing data.
- Data profiling: Use data profiling techniques to extract missing data.

Types of imputation:
- Mean imputation: Replace missing values with the mean of the variable.
- Regression imputation: Replace missing values using a regression model.
- Multiple imputation: Create multiple versions of the data with different imputations.
Imputation methods:
- Hot deck imputation: Replace missing values with values from a similar respondent.
- Cold deck imputation: Replace missing values with values from a different data source.
- Predictive mean matching: Impute missing values using a predictive model.

There are three types of missing data: MCAR (Missing Completely At Random), MAR (Missing At Random), and NMAR (Not Missing At Random).
Estimation methods include Listwise deletion, Pairwise deletion, and Regression imputation.

Data errors can occur in three forms: Data entry errors, Measurement errors, and Data processing errors.
Data cleaning techniques include Handling outliers, Data normalization, and Data transformation.

There are two types of missing data extraction: List extraction and Pair extraction.
Extraction methods include using SQL queries and Data profiling techniques.

There are three types of imputation: Mean imputation, Regression imputation, and Multiple imputation.
Imputation methods include Hot deck imputation, Cold deck imputation, and Predictive mean matching.

Logistic regression predicts a binary outcome (0/1, yes/no, etc.) based on one or more predictor variables
Estimates the probability of the outcome, with the goal of finding the best fitting model for accurate prediction

Measures the strength of association between a predictor variable and the outcome
Represents the change in odds of the outcome occurring when the predictor variable increases by one unit
Interpreted as the change in odds of the outcome for a one-unit change in the predictor variable, while holding all other predictor variables constant
OR > 1 indicates a positive association, OR < 1 indicates a negative association, and OR = 1 indicates no association

Graphically represents the model's performance, plotting true positive rate (sensitivity) against false positive rate (1 - specificity) at different threshold settings
Evaluates the model's ability to distinguish between positive and negative outcomes
Area under the ROC curve (AUC) measures the model's overall performance, with higher values indicating better performance

Exponent(B) (e^B) is the odds ratio, which represents the change in odds of the outcome
e^B is always greater than 0, and the larger the value, the greater the change in odds of the outcome
Used to interpret the results in terms of the odds ratio, which is more intuitive than the log-odds