Levels of Control in Data Gathering

In scientific data gathering, there are three different ways to find the data needed to look at the relationships being studied. These three methods will tell you different things, and have their own strengths and weaknesses. The three methods are the experimental, the quasi experimental, and the correlation method.

In the experimental method, the scientist is trying to isolate causation from a single variable.  To see if a particular variable has a consequence the scientist must hold all other variable to be the same and only change the one thing that they are studying. To make sure that the experiment is being done correctly, the choice of who is exposed to the independent variable must be at random (Salkind 2014 p. 8). For example, say that the scientist thinks that college men wearing a baseball cap will be able to run faster. The scientist will then distribute baseball caps to the study group and then measure the running speeds of all the participants. If the study finds that the average speed of the baseball cap wears was in fact faster than the average speed of the non-cap wearers, then the hypothesis is confirmed.

Not all variables are as easily tested as the question of speed of college-age men and baseball cap wearing. Sometimes what is to be measured is not so easy to control.  If the experiment designer cannot pick who receives a variable, then there is a measure of control lost. This becomes part of a quasi-experimental method (Salkind 2014 p. 9). An example where a quasi-experimental method would be used is one where a scientist was curious at who was better at chess, all other things being equal, left-handed people or right-handed people. Since nature has already chosen who will be left-handed and who will be right handed, the element of randomness has been taken away from the experimenter. The ultimate results of the quasi-experiment may be less certain than a strictly experimental method because there may be other variables that the left-handers possess other than their dominate hand that may skew the results.

The final method for looking at relationships between two variables is the correlational method. In this method, there is no experiment run, but the scientist looks at two sets of data to see if there is a relationship between them. Does one go up while at the same time the other goes down? Alternatively, do they move in tandem together? If they do either one, then the indicators are said to be correlated. The problem with looking for correlation is that scientist cannot tell if there is a direct causal relationship (Salkind 2014 p. 10). Say a scientist can look at the sales of Happy Meals in America as well as the average weight of American children. If the scientist sees that both variables increased over the same time, then a correlation can be said to exist. The issue is that there is no way to say directly what caused what. Did children gain weight because they were eating too many Happy Meals, or did already-obese children demand more Happy Meals?

The overall result is that the more control a scientist has over the independent variables that they are studying, the more certain they can be with the validity of their results. In the use of data, more control is the desired starting point, but it may not always be possible to attain. That is why the other options exist.

Bertrand Russell Deserves a Seat at Your Table: On Sceptical Essays

I grabbed this book because it was in the Journal’s recommend books for year-end last year. I had read his “Why I am not a Christian,” and was aware of Russell as a philosopher and mathematician. I did not know he was such a clear writer. I have to respect a free  thinking, socialist, atheist from 100 years ago who was not afraid to follow the strength of his convictions even though they led him against the grain. He lost potential jobs, and went to jail for his beliefs. Maybe he was never in any real danger, but I don’t know – still brave.

Reading this book made me think of that hypothetical situation where you can have a dinner with anyone you want, living or dead. I think I’d have Russell at my table. His writing, reading it now, sounds contemporary.  These essays, for the most part, would not be out of place in current conversation. I say for the most part, because there are a couple that strike wrong notes. One essentializes all “Chinese,” the other talks about the benefits of behavorialism and is perhaps too enthuastical about the problems that science could solve. Other than that, I liked all the essays. In fact, I liked them so much that it is hard to point out what was good. I normally read with a pen so I can take notes and engage with the text, but I couldn’t with this book. It just had narrative and argumentative momentum that I couldn’t dent. I instead dog-eared the pages where there was a striking turn of phrase of interesting way of looking at a subject that I hadn’t previously considered. By the end of the book, my wife remarked at just how many dog-ears were in the book. I can’t summarize it here and give it justices. You need to read Russell to appreciate him. I’m just a shadow on the cave wall.

Median Versus Mean

In my experience, there is a big difference in when the median or the mean is the most useful tool to look at when you want to find the central tendency.  The question is what the likelihood of vast outliers is.

A good example is height. The human body can only grow so tall. That means that if you want to find out what an “average” person is in height, then the arithmetic mean is useful. You take a large enough sample, and then you can find with some large degree of certainty that the average height of the population is close the arithmetic mean of your sample. Most likely it will also be close to your median height, as the standard distribution was built on such measurements.

Conversely, some measurements are not bound. An example here is wealth. There is no biological, physical, or chemical reason that a person cannot have all the money in the world. There are people in the world who have so much money that most people in the world could not imagine it. There are then also people who make that first person feel poor. This long tail can lead to some misleading measurements. If you took the wealth of Bill Gates and three random people off the street, the mean wealth of your sample would be around 20 billion dollars. That may sound absurd, but some people have so much money that they can have a heavy weight on the mean of the whole population. That is why economists like to talk about median household income. Currently, if I recall correctly, it is around fifty thousand dollars. The mean household income is much more than that. We use the median here because it gives us a truer picture of the thing being measured.