Using the preceding plots and missing value charts, it was easy to figure out the count, percentage, and spread of missing values in the datasets. We noticed that many variables had missing values for the same observations. However, after consulting the data description, we saw that most of the missing values were actually not missing, but since they were codified as NA, pandas treated them as missing values.

It is very important for data analysts to understand data descriptions and treat the missing values appropriately. 

Usually, missing data is categorized into three categories:

• Missing completely at random (MCAR): MCAR denotes that the missing values have nothing to do with the object being studied. In other words, data is MCAR when the probability of missing data on a variable is not related to other measured variables or to the values themselves. An example of this could be, for instance, the age of certain respondents to a survey not being recorded, purely by chance.
• Missing at random (MAR): The name MAR is a little misleading here because the absence of values is not random in this case. Data is MAR if its absence is related to other observed variables, but not to the underlying values of the data itself. For example, when we collect data from customers, rich customers are less likely to disclose their income than their other counterparts, resulting in MAR data.
• Missing not at random (MNAR): Data is MNAR, also known as non-ignorable if it can't be classified as MCAR nor MAR. For example, perhaps some consumers don't want to share their age when it is above 40 because they would like to hide it.

There are various strategies that can be applied to impute the missing values, as listed here:

• Source the missing data
• Leave out incomplete observations
• Replace missing data with an estimate, such as a mean or a median
• Estimate the missing data from other variables in the dataset