types of missing data in research

Types of Missing Data in Research: A Comprehensive Guide

By Melissa Harris
AD

Missing data is a common challenge in research that can significantly impact the validity of results. It occurs when there is an incomplete or inaccessible part of the data that was supposed to be collected. Understanding the types of missing data is crucial for developing appropriate strategies to handle them. This article will explore the different types of missing data in research.

Missing data can have various causes and patterns, which can be categorized into different types. According to Little and Rubin's (2002) classification, the most common types of missing data in research are Missing Completely at Random (MCAR), Missing at Random (MAR), and Not Missing at Random (NMAR).

Missing Completely at Random (MCAR)

MCAR is the most benign form of missing data. In this scenario, the probability of data being missing is unrelated to the observed or unobserved data. In other words, the process that generates the missing data is completely random and does not depend on any part of the observed or unobserved data. For example, if a survey is conducted, and the missing responses are due to random factors such as power outages or Alianzaanufacturing](//www.alianzaan.com), then the data is MCAR.

AD
Missing Data in Time-Series: Machine Learning Techniques | Towards Data Science
Missing Data in Time-Series: Machine Learning Techniques | Towards Data Science
a flow diagram with the words handling missing data in different colors and shapes, including arrows
a flow diagram with the words handling missing data in different colors and shapes, including arrows
Data Collection Methodology Guide, Quality Data Collection Methods, Data Collection Methods In Education, Data Collection Methods Overview, Quantitative Data Collection Techniques, Quantitative Data Collection Methods, Data Collection Methods In Research, Data Collection Methods, Qualitative Data Collection Techniques
Data Collection Methodology Guide, Quality Data Collection Methods, Data Collection Methods In Education, Data Collection Methods Overview, Quantitative Data Collection Techniques, Quantitative Data Collection Methods, Data Collection Methods In Research, Data Collection Methods, Qualitative Data Collection Techniques

To illustrate, let's consider a clinical trial where participants' weights are recorded. If the weight data for some participants is missing due to a power cut during the weighing process, it can be considered MCAR. The probability of missing data here is unrelated to the participants' weights, observed or unobserved.

Imputation Techniques for MCAR

Imputation is the process of filling in missing data with some estimated value. For MCAR, simple imputation techniques like mean, median, or mode imputation can be used. These methods replace the missing value with the mean, median, or mode of the observed values of that variable respectively.

a table that has different types of data and information on it, with the text below
a table that has different types of data and information on it, with the text below
Missing Data Overview: Types, Implications & Handling
Missing Data Overview: Types, Implications & Handling
Missing Data Conundrum: Exploration and Imputation Techniques
Missing Data Conundrum: Exploration and Imputation Techniques

For instance, if the mean weight of the observed data is 70 kg, then the missing weight values can be replaced with 70 kg. Although these methods are simple and easy to implement, they may not capture the true complexity of the data and can introduce bias.

Missing at Random (MAR)

MAR is a less benign form of missing data compared to MCAR. In this case, the probability of data being missing depends on the observed data but not on the unobserved data. In other words, the process that generates the missing data is related to the observed data, but not the unobserved data.

AD

For example, in a survey about income, if the reason for missing data is that respondents declined to provide their income because it's too high (a variable that is observed but does not directly relate to the unobserved income), then the data is MAR.

a flow diagram showing the steps to writing a research paper
a flow diagram showing the steps to writing a research paper
Data Collection Methods
Data Collection Methods
a man in white lab coat looking at laptop screen with data on it and the caption says, confused with your research data? don't
a man in white lab coat looking at laptop screen with data on it and the caption says, confused with your research data? don't
the word data is shown in red and black text, with other words surrounding it
the word data is shown in red and black text, with other words surrounding it
Different types of missing data.
Different types of missing data.
📉 Common Data Analysis Mistakes Beginners Make
📉 Common Data Analysis Mistakes Beginners Make

Model-Based Approaches for MAR

To handle MAR data, model-based approaches like the Expectation-Maximization (EM) algorithm or multiple imputation (MI) can be used. These methods assume that the observed and unobserved data come from a joint distribution and can provide valid inferences about the unobserved data.

The EM algorithm alternates between estimating the likelihood of the observed data and estimating the Expectation of the complete data. Multiple imputation, on the other hand, creates multiple imputed datasets and then combines the results to provide valid inferences.

AD

Pattern-Mixture Models for MAR

Pattern-mixture models are another approach to handle MAR data. This method combines the strengths of the two approaches mentioned above. It examines the relationship between the probability of missingness and the outcomes, as well as the lack of a relationship between them.

the data collection method is shown in this diagram, which shows what it looks like to be
the data collection method is shown in this diagram, which shows what it looks like to be

For instance, if we observe a pattern where individuals with higher levels of education are more likely to have missing income data, the pattern-mixture model can account for this pattern of missingness and provide a more accurate estimate of the income levels.

Not Missing at Random (NMAR)

NMAR is the most complex and problematic form of missing data. In this case, the probability of data being missing depends on the unobserved data. Therefore, the process that generates the missing data is related to the unobserved data, which makes it difficult to infer about the unobserved data.

AD

For example, in a study about the effectiveness of a new drug, if the patients who experienced severe side effects (unobserved data) refused to continue the study (missing data), then the data is NMAR.

Logistic Regression for NMAR

To handle NMAR data, logistic regression is often used. Logistic regression models the probability of missingness as a function of observed and unobserved data. The key assumption here is that the probability of missingness only depends on the observed and unobserved data, not on the unobserved missing data.

For instance, in the aforementioned drug study, logistic regression can model the probability of missing data (i.e., patients dropping out) as a function of observed data (e.g., patients' demographic information) and unobserved data (e.g., patients' initial health status).

Understanding the types of missing data is the first crucial step in developing appropriate strategies to handle them. By doing so, researchers can minimize bias and improve the accuracy of their findings. As such, the field of statistics and methodologies for handling missing data continue to evolve, providing us with more sophisticated tools to address these challenges.

AD