Question: How do you design your experiments?

Anna Williams answered on 24 Jun 2011:

Hi 10msvon,
This is a really good and quite difficult question!
It really depends on my what I am trying to find out, and what my hypothesis is.

I will give you an example from a project that I am supervising at the moment, which has a simple experiment in it. We are trying to find out how decomposition affects the mineral part of bone. We cannot use humans (that would be a bit unethical!), so we use pigs, as they are very similar to humans in their internal anatomy.

We have designed an experiment where we are looking at 4 pigs decomposing. We took a bit of bone from each one when they had just died. This acts as the ‘control’ because this sample does not undergo decomposition, so we know what the bone mineral is like without decomposition. We need this so that when we have look at the bone that has undergone decomposition, we can compare it to the sample without decomposition. If there are any differences between the two, then we know that those differences are caused by the decomposition, because that is the only thing that has happened to one that hasn’t happened to the other.

Then we allow the pigs to decompose. We make sure that the decomposition is a natural as possible, because we want the results of this experiment to be applicable and relevant to normal situations. After the decomposition has finished (or got far enough), we take some bone from the pigs again, and look at the bone mineral. We have to make sure that we take several samples, and do the test to look at the bone mineral several times. This is to make sure we get the same answer again and again. This is called doing ‘repeats’ to ‘reduce error’. This means that the answer we get, if we get it again and again, is the right answer, and not just a one time fluke. We have chosen 4 pigs so that we get a lot of samples to repeat the tests on. This is because one pig could decompose in a funny way or have funny bones that were not representative of pig bone in general. We need several pigs to make sure that the results we get are more representative of pigs in general. We should get lots of samples and lots of the same answer again and again.

Here is a really good introduction to designing experiments: http://www.longwood.edu/cleanva/images/sec6.designexperiment.pdf