To learn a statistical package?

[Genecks]

I was told by some people that I should learn a statistical package:

http://en.wikipedia.org/wiki/List_of_statistical_packages

 

Anyone know why I should? Which one is the best?

[layzwombat]

Basically any analysis requires statistical evidence. Depending on the data, you have to run a certain set of tests if say you have 3+ samples. You absolutely can do these by hand, if you have the time and patience, but a statistical package will make this much easier. SAS and SPSS are fairly common in my area (SPSS being much more user friendly), but I've noticed that text books like minitab - and I'm not sure why.

[Bignose]

Well, another option between doing it by hand and using a package is to program the analysis yourself. The drawback to this is time and debugging your own code, and you need to be familiar with computer programming language and the statistical formulas and their appropriate use.

 

The real benefit is that you know exactly what the computer is doing -- too often these packages are treated as balck boxes where you do not know what the computer is doing to your data. The packages will allow you to do anything you can click your mouse on, but the old computer adage remains very important "Garbage In, Garbage Out"

 

What I mean by this is that sure you can perform Test D5 (just made that name up as a fake example!) with SAS on your data, and the computer will give you an output, but was the test even appropriate? Do you have the right data for that test? Enough data? Enough to be meaningful?

 

If you have a good knowledge of the formula and the theory behind the test you can answer these questions yourself rather than just taking the computer's word for it. The best example of what I am talking about is the over use and abuse of the R^2 statistic. All too often, way too much emphasis is placed on this statistic when in reality its usefulness and information content is very, very low.

 

All that said, the commericial packages have been written to make the tasks easier and faster. If you are going to have to do tens, hundred, thousands of statistical tests, you may not want to invest the time to write your own program for each one of those. The real question that needs answering here is what are your needs going to be?

[layzwombat]

Another thing to note about the statistical packages is what bignose stated: bottom line, if you do not know what a test is doing (ie. test of location - mean or median, variance, etc) you probably won't know the assumptions required for the test to be run in the first place. That being said - if you do, most programs will give you the proper information in the output as long as you know what to look for.

 

Example: If you run an ANOVA test (equal variance assumption - others should be worked out in the study design) SPSS will use a Levene statistic to test before it gives you the results from the ANOVA - if Levene fails, than obviously the ANOVA is irrelevant, and you have unequal variances (Try a Welches ANOVA in this case). I believe SAS uses a Folded-F test to do the same thing.

 

In my opinion the understanding a statistical package is a great compliment to the skill, and it looks good on a resume. I would have to recommend SPSS as it is becoming more common, and easier to use than most packages. It also has an interface similar to an excel spreadsheet.

[Glider]

I was told by some people that I should learn a statistical package:

http://en.wikipedia.org/wiki/List_of_statistical_packages

 

Anyone know why I should? Which one is the best?

I don't think there is a 'best', sisnce many are designed for different purposes (e.g. market research, epidemiology/population studies etc.).

 

One of the most comprehensive is SPSS (statistical Package for the Social Sciences). This one is good because it is flexible. You can buy the basic package and then add different modules and functions to it according to what you need it for.

 

However, as has been said, if you don't understand the basics, then none of them will be much help. Learning to use a statistical spreadsheet is different from learning to use statistics in the same way that learning how to use all the tools in a toolkit is different to learning how to strip and re-build an engine.