Experimental Design – Correlative or Manipulative

A correlative study is where you look for associations between an observation and one or more other observations; eg to test “Smokers have lower lung capacity than non-smokers.”

A manipulative study is where you deliberately change something to determine if it has an effect eg to test “lung capacity increases after stopping smoking.”

When are you most likely to choose correlative studies?

  1. Situations where manipulation is impossible (practical or ethical reasons)
  2. As pilot data (observational data) to establish hypotheses that you subsequently will test using manipulative approaches

When would you choose manipulative studies?

  1. Where you need to control for third variables/confounding variables
  2. Where reverse causation is a concern
  3. Whenever mechanistic insight is required

Key point; Good papers will often contain both correlative and manipulative experiments within their story. Correlative studies are often used to establish a real world connection and generate a hypothesis, manipulative studies will then test that hypothesis in a controlled, closed system.

If you  know what you are going to use then head on to MEASUREMENTS, if you want to know more about what you should think about in making your decision then read on.

Correlative or Manipulative

Key point: Is manipulation possible?

Usually (not always) manipulative studies will allow you to use stronger statistical tests and will yield easier to interpret results. They are the approach of choice for most bench science. However, there are lots of times when it is practically infeasible or impossible to actually manipulate the system in the way that you want. Lots of examples here, but the most obvious are when you are studying a long-term outcome or when it is ethically inappropriate to intervene (take the smoking example above; getting people to start smoking in their 20s so you can see if they get cancer in their 50s is both ethically unforgivable and practically impossible).

Correlational studies can often (not always!) be quicker, easier and less invasiveEssentially you need only to gather your study population and make observations and so it is often prudent to do a correlational/observational study as a pilot study.

Key point: Can you produce biologically realistic manipulations?

Screen Shot 2018-03-21 at 20.01.57Correlation studies also benefit in that you are studying true biologically relevant variation – looking at the real life scenario will give you the opportunity to study whatever you care about in the context of everything else that might affect it. In terms of choosing a manipulation, are you might be asking your subject to do something that they wouldn’t normally ever experience. This works at the cell level as well as the whole organism; if, for example, you are driving overexpression of a protein with a viral promoter you may end up with a level of expression that would never happen in real life so you will be limited in how you interpret the data.

How do you decide if your manipulation is realistic? The answer likely depends on your question and specific hypothesis – eg in the overexpression experiment I just described you can determine if your protein is capable of interacting with other proteins but you would have to be cautious about interpreting your data as your protein does interact.

At this point, you might need to go back to your questions/hypotheses and adjust them so they are appropriate for what you are actually able to test. Remember your hypothesis has to be testable!

Key point: Will your manipulation lead to any non-intended effects?

They’re non-intended effects so how will you know?! This is where your knowledge of biology and common sense come in again. Questions you should be thinking about here are things like off-target effects of your treatment be it a drug, miRNA, energy drink or whatever. We will try to account for these issues by using control treatments (more on this later )  but if you are unable to design out these problems then you might not be able to answer your question and may need to backtrack a little in your thinking.

Key pointCorrelation does not imply causation 

You’ve probably heard this a 1000 times! The biggest weakness of correlative studies is that you can’t be sure if the two observations are actually linked or if there is some other reason for observing similar trends. You will continue to think about this point throughout the rest of your experimental design to try and maximise the probability that the things you observe are indeed linked; this will be done by choosing appropriate groupings, and controlling for confounding variables also known as third variables. More on this later.

Spurious correlations… Meaningless? or is it?

chart via http://www.tylervigen.com/spurious-correlations

Key pointCorrelation does not indicate directionality 

Does A influence B or is it really that B influences A? This is called reverse causation and again it’s something that you have to consider in the design of your experiment. If you can’t design out the reverse causation problem then a correlative study may not allow you to answer your question.

Example; body mass index is inversely correlated with number of minutes of exercise per week but does this mean people with higher BMI exercise less or that those people who exercise less have higher BMIs? Just measuring BMI and activity in this case won’t tell you the direction so will limit your interpretation. 

Ok, you’ve decided the type of study, it’s time to think about what you are actually going to measure. MEASUREMENTS.