Types of Missingness

Today, we'll discuss the first type of missing data: Missing Completely At Random, or MCAR. This occurs when the likelihood of a data point being missing is completely unrelated to any other information. Can anyone give an example of this?

Maybe when someone accidentally skips a question on a survey?

Or when there's a glitch in data collection that causes random rows to be missing?

Exactly! MCAR has no bias on statistical inference, which is why it’s less problematic. Remember the acronym: MCAR—Missing Completely At Random. It helps in retaining data quality! Why is it important to detect whether data is MCAR?

So we know how to handle it correctly?

That's right! Recognizing MCAR allows you to decide whether to delete missing entries without worries. Let’s summarize: MCAR means data is missing randomly, not impacting our results.

Next, we move to Missing At Random, or MAR. In this case, the missingness can be explained by other observed variables. Any examples?

Perhaps if older individuals tend to skip questions about technology on a survey about digital habits?

Or if people don't answer because they might feel embarrassed about their income?

Yes! Although this is more complex, it's still manageable. MAR implies that we might still accurately infer the missing data using the information we have. Remember: MAR—Missing At Random. Why is this concept critical in data analysis?

Because we need to use appropriate approaches, like imputation, to reduce bias.

Correct! It leads to better performance in predictive modeling.

Now, let’s discuss Missing Not At Random, or MNAR. This is the trickiest type, as the missingness reflects the value of the missing data itself. Can you think of situations where MNAR might apply?

If someone with a high income did not want to disclose their income due to privacy concerns?

Or in health surveys where individuals with severe symptoms avoid answering certain questions?

Exactly! MNAR makes it difficult to analyze data accurately because the missing values are non-randomly distributed. Remember: MNAR—Missing Not At Random. How should we deal with MNAR cases?

We may need to employ more advanced techniques or accept that it's an intrinsic bias in our data.

Right! MNAR must be treated carefully because it can lead to substantial bias in results. Let's recap: MNAR relates the missingness to the unseen data, leading to potential information loss.