Now that the analyses have been run, we have context on what the data represents. You can still write any additional analyses, but you need to be careful how to report them. So, let's take a moment to separate the two kinds of analyses.
This is a recap of confirmatory vs exploratory analyses that was briefly mentioned in previous blog posts.
Confirmatory vs Exploratory
There are two types of analyses that can be performed on any dataset. The analyses themselves can be anything; however, they differ depending on context and the current part of the study process.
Context depends on how much information you currently know about your dataset. This could range from simple statistics like the mean or mode to complicated regression or detection models. Confirmatory analyses are analyses where you have as little context as possible about the dataset. You are confirming the methodology of your study, so the more you know about the dataset, the less you need to confirm about the initial methodology. Exploratory analyses happen when you know the full context of the dataset. In this case, we are using 'full' to describe whatever you are initially measuring. Any analyses made after you have context are biased, so we may have some confounders which might not accurately represent the data.
The study process can also affect whether your analyses are confirmatory or exploratory. Usually, this is subjective depending on the researcher; however, we can provide some bounds. All analyses written before a preregistration is published are confirmatory since preregistrations hold the methodology of a study before the study runs. Likewise, all analyses after an analysis has been run on the dataset are exploratory, similar to the reasons above. A general midpoint most researchers will agree on is that any analyses decided on before viewing the data are confirmatory while all after are exploratory. This is because viewing the data is the first context the researcher may get to a study, which also provides the first form of bias a researcher may have.
Reporting Exploratory Analyses
Reporting exploratory analyses is relatively simple: you just need to state that the analyses are exploratory when reporting. This is typically within the same paragraph of the reported results, or it could be put into a separate subsection. If you decide to include the exploratory analyses in the methodology section, the analyses themselves should be prefixed with some sort of disclaimer.
Some Additional Thoughts
A lot of research uses exploratory analyses
A number of research articles contain exploratory analyses, regardless of whether they are actually exploratory. This typically falls down to two reasons. First, there are a number of research articles which don't report whether their analyses are confirmatory or exploratory, so we take the looser interpretation. Second, research is full of accidental p-hacking or harking because of the lack of documentation on the original analyses. In general, we just need to make sure we document everything such that our actual confirmatory analyses can be trusted.