Monday, February 28, 2011

My analysis report or How to analyze large amounts of test data

The past two weeks and today, I've been continuously working on my analysis report. Today, just before lunch, my first draft (which still needs introduction and conclusions) was rolling out of the printer. I thought it was not so bad how I put together a 23k words document and a large number of graphs and tables in only 11 days' time (including the time to make the analysis calculations).

I wasn't in my best shape in terms of concentration (my housing situation has cause me a lot of tension lately, although the final solution is very near), so I've been randomly clicking around the interwebs too much. However, I still think my workflow with regard to making this document was neatly organized.

Only a few months ago, I spent about the same amount of days on an analysis document of less than half the size of this one, and I spent all those days working until 8pm (instead of 5 - 6pm as I'm trying to do now).

So here's how I organized this project:

1. Review what is already done.

I already had two previous versions of the analysis report of my experimental work, so I could recycle at least some material. I ended up mainly recycling the style of my graphs, and reanalyzing my data since my new test results were screaming to be included in every possible subchapter. I also knew my advisor had suggested changing the amount of data points I'm squeezing into the plots. I waited until this version to make this change.

2. Check the expectations.

Before starting (or better, before realizing another deadline in our project was coming up), I had an appointment with my daily supervisor who told me very clear which questions our funder wants to be answered. I really appreciated that I knew exactly in advance what the expectations for this report are, so I can work towards this in as much as I can.

3. Know what needs to be done.

So I knew what the expectations were. I printed out a previous version of the report and started to reread it. In every section I jotted down what had to be altered, recalculated or added. I also changed the order of the sections, and ended up with a document filled with pencil-scribbling to guide me what had to be done, section by section. I started off with putting the sections in my word document into the right order, and then I could get started altering, improving and adding material to the sections.

4. Give the reader some framework.

Previously, I just referred to the reports with the test data. I now added a short chapter with a sketch of the test setup and some basic information about the specimens. I think that was a good idea.

5. Study the parameters as resulting from the tests.

And so I've spent 11 days playing around with data in Excel and having rows and rows and columns and columns of data flying around before my eyes. My first action was to filter my data into nice plots and tables per parameter which we had been testing.

6. Compare to calculation methods in general.

My second action then was to compare the experimental values to the calculated values. Turn out, I had to recalculate most of my calculated values as I had only made rough predictions before the tests, and I did not implement the properties as we had measured them on the test data. Luckily, I have a nice set of MathCad sheets that do the job for me. Also, once I had a spreadsheet set up to analyze my data through one method, using the next method was only a matter of copy and paste.

7. Compare how the studied parameters are reflected by these methods.

Since the comparison with the Dutch code as calculation method gave me quite disappointing results, I decided to go and check per parameter the comparison of test value to calculated value to see where exactly the weak spots of the methods are. Doing that, I did some nice observations, which motivated me to play around some more with all my data.

8. Give recommendations.

It doesn't really help the fund to produce graphs and tables and let them look for the answers to their questions in all the material I've produced. So today, I printed out the document, started reading from the beginning, and jotted down the most important recommendations - per calculation method and in general.

9. Review your material.

Does it make sense? When reviewing my material today, I found that one of my tables raised my eyebrows. I went back to check it, doublechecked my calculations, and then I discovered that I had been reading results from the wrong column.

How do you analyze large amounts of test data? Any advice for me? I'm just hoping to give my funder a report which gives them as clear as possible the answers to their questions, but also gives them enough background to these answers.

1 comment:

  1. Hi Eva,
    Great blog and great to see you are using Mathcad. I am currently looking for Mathcad users to run quick interviews for publishing on the Mathcad blog -
    If interested, pls reach out to me at bgiemsa @ ptc dot com - Thanks!