The End of Tables? Of Confidence Intervals?

I have long found the quantitative side of political science research deeply fascinating. The various estimation methods, sampling distributions and arguments about Bayesian vs Frequentist approaches amuse me to no end. The more detail provided about methodology and analysis in an article the better. However, I also recognize that my deep enjoyment of these aspects of political science make me a bit of an outlier (at least in some parts of Europe!).

A friend of mine who also happens to be a PhD student at UCD (link) and happens to study electoral behavior and turnout has spent many an hour trying to show me the light; that our research is at its best when it can be conveyed to the “real people”. It has been a difficult education for me, but I’m slowly coming around to a position in which I agree that, for this field, there should always be something of value that can be taken from our research and conveyed in a clear and interesting way to people who don’t care that my OLS is BLUE (I know, they make me sad too).

Obviously I am slightly exaggerating my position. As my research has often been rooted in the politics of the environment it is nearly impossible to separate out either a normative or real-world aim to what I’m doing.  In fact, I have broad plans for the “real-world” applicability of my research, but I leave that for another time. Nonetheless, our conversations have been most useful in helping me design my graphs more clearly. As another friend (link) described him, he has an Andrew Gelman-like ability to nit-pick any graph to pieces. I’ll steal the concept, creative destruction, to describe his talent.

Kastellec and Leoni have an article which is very much on point for these discussion from an actual practice perspective. They detail an interesting way to replace simple and complex tables with concise graphs. An excellent read, and one which I have taken to heart and incorporated into my methods. While I think they have demonstrated an incredibly useful technique for displaying table results clearly with coefficients and confidence intervals, I’m not sure my friend has been sold. If I wanted to completely misrepresent his position to provoke a response I would characterize his position as “if it’s not a barplot I don’t want to see it”.

I can imagine this debate never being completely settled which is, I think, a good thing. Having people you respect constantly pushing you to provide greater clarity can only strengthen your argument, and hopefully, I can eventually convince him that without confidence intervals we have nothing…

Below is a very basic example with invented data just to convey the idea:

(The points on the graph are the coefficients and the solid line represents the 95% confidence interval. Obviously, when the CI includes the vertical line at zero, the null, that Beta_i equals zero, cannot be excluded.)

Graphical Representation

Graphical Representation

.

.

Kastellec, J.P. & Leoni, E.L., 2007. Using Graphs Instead of Tables in Political Science. Perspectives on Politics, 5(04), p.755-771.

Advertisements

One response to “The End of Tables? Of Confidence Intervals?

  1. Stephen Quinlan

    Firstly, I want to congratulate you for beginning this useful discussion – it is always a good thing that political researchers/scientists look at the various ways of presenting their results.

    I just want to present my position clearly because it was somewhat misrepresented (I don’t take offence – that is after all the American way). The main premise of my argument is that the presentation of results depends on a number of factors. Firstly, the audience you are presenting to. If it is fellow political scientists, I think the graphical/tabular presentation is not as sensitive an issue as it is likely that if they are presented using conventional techniques (whether it be tables or figures), it is most likely that our colleagues (well those of them in the quantitative side of things) will understand. However, my point is that when we are presenting our results to ‘real people’ (by that I mean people who do not have a degree in political science), we should try and present our results in as simple a way as possible. Now, of course this all depends on what you are trying to explain and the methods you used to come to your conclusions. The reality is that some methods are extremely complicated and the results of which can only be conveyed to an audience by going into details which may be difficult for them to comprehend. But in a lot of instances this is not the case. It is for the latter that I advocate taking a simple approach to presentation of our findings. The simple truth is a broader audience is able to comprehend bar plots, line graphs etc… If it can be presented in this way and we are trying to promote our research to the man on the street, political scientists have to be aware that if it can be conveyed in a simpler fashion, it should be conveyed in a simpler fashion. There is no point in going for the complicated explanation (unless it is necessary, a point which I have acknowledged earlier) and showing how smart we are if the majority of people leave the room asking each other ‘what the hell was s/he talking about?’ In sum, if we can achieve it simplicity is best when conveying our results to people on the street. After all, we want our results to be conveyed to the broadest possible spectrum…

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s