Last 2 weeks in April were crazy for me with finals and term paper submissions and assignments and trying to wind up experiments. As far as research was concerned (my 2nd rotation in an awesome lab), I was able to stick to the plan I made - did all the experiments that I had to and was left with data analysis in the last few days. Yes, the other side of things. Getting experiments to work is just the first half of the story; trying to make (any) sense of the data you get out of experiments is the other half.
It was only after I plunged into this other half, did I realize how much of data I generated that it felt close to impossible to analyse all of it before my big day – when I’d be presenting in the lab meeting. I was beginning to wonder which is the tougher one that makes you go crazy – the experiments or the data analysis. It really tires you out. Though I got through everything and even got some interesting results, I still don’t have an answer to that question (which I think is a profound one).
But I did learn a couple of things about this whole process of studying data. The most important thing being - never leave this task to the end. Never. The best way to go about it is to start getting a hang of how the data looks as and when you have some of it ready. In my case, each time I did an experiment I took a good number of movies so I could have a good number of data points but once I started analysing the movies, I realized that they were not enough. I needed so many more movies. This happens when you are studying motors in invitro assays (which was what I was studying). There were also some data sets where I felt I’d be better off with a longer movie, maybe slower/faster frame rate and such things. Its always good to know what are the things you need to work on before you do the experiment again at a later point.
Another important aspect – honesty. Since I sort of knew what kind of a trend or pattern I should see in the parameters that I was studying across several different experiments, each time I got to a point where I was able to compare these cases, my mind would try to work its way and make sure I was seeing exactly the expected trend. I found it slightly hard in the beginning to do a completely unbiased analysis. But I got around this by telling myself that what people have seen in the past need not be true at all and maybe I’d see something novel (well, I know this is not true in most cases but you get what I mean, atleast it motivated me to do a blindfold analysis). Its very important to be completely honest with yourself and be able to communicate to others how exactly you went about analysing your data. That way me giving a talk in the lab meeting was good – I realized each person has a different style of looking at his/her data and even representing it. I think that’s one of the things that has to be given a good thought – what is the one best way to represent your data that will definitely make people think and not just listen.
All said, I’m glad I learnt these small things early on. Also, just so you know my talk went very well and the rotation was extremely satisfying :)
PS: About the title - I don't think there's an ugly side to data analysis, is there?
Quite agree with the data analysis point. Something one also sees with theoretical work. Sometimes if one doesn't do atleast a preliminary analysis, one even forgets contexts and it gets irritating. One has to do a good book-keeping! A difficult habit to get into..
ReplyDeleteThe honesty - yes, this thing of being prejudiced is very difficult to get out of. One is very badly prejudiced sometimes and one starts to think some of the results we obtain are simply wrong if they don't agree with 'established' results (sometimes they are 'established' only in our heads!) and in the process, lose some useful interpretations. Taking things at face value, but at the same time, trying to see the whole picture - a delicate balance one has to achieve, no wonder, they say sometimes science is still an art, in some measure..
Sathej
@Sathej - You put it so well and I like what you said about 'established' only in our heads. That feels so true.
ReplyDeleteSince I know what kind of data you looked at, I think it was not really wise of you to do all that manually. You should have written (or get) a code (since you claim to know python) to analyse all that data without any human intervention which can address both the issues you mentioned here - the time required to analyze the data and the bias in interpreting it.
ReplyDelete@Rk - I definitely could've done that and then this post would never have been written but yes, I'm with you on this. I think thats the best way to go.
ReplyDeleteThis comment has been removed by the author.
ReplyDelete