Analyse-it blog The latest news on new features and software releases.

8-Jul-2008 Choosing the correct statistical test

Many Analyse-it users readily admit their statistics knowledge is a little rusty, usually because it’s 10 years or more since their last statistics course. Should I use the t-test, Mann-Whitney, or Wilcoxon test? The names of the tests aren’t exactly helpful, nor do they give you any clue of the assumptions that must be met to use the test.

That’s why we devised the Statistical Test Advisor.

It’s a simple interactive wizard that asks what you want to do, what data you’ve observed, checks which pre-test assumptions can be met, then tells you the best statistical test to use. Using the advisor you can be confident you’re using the correct statistical test -- or even use it to check if your statistics knowledge really is as rusty as you think!

Try it for yourself now:

/support/advisor/test-advisor.aspx

Bear in mind this is only a simple prototype at the moment. Eventually the advisor will integrate into Analyse-it, leading you to help and tutorials showing you how to use the recommended test and interpret the statistics. You’ll also notice the advisor recommends tests that will be new to Analyse-it 3.

So now it’s over to you. What do you think to the test advisor? Will you find this useful? Are the questions logical? Is the terminology clear or do we need to explain the terms used? Finally, have we missed anything you think should be included?

Let us know what you think, good or bad. Post your comments below.

Comments

For a first cut I think it is pretty good. If it were to become a little more interactive it would be better. But for the type of people you have described (like me, no stats class for over five years) I found it to be a great idea. Thanks.
The screen description does not show the full name of the test at final decision description screen. All I see is dots(.........)
Thanks for your comment Don. Can you elaborate on how we could make the advisor more interactive?
Probably 20% of my year is involved with gathering data and statistical analyses the remaining 80% is dealing with the conclusions. The most frustrating part of books I have purchased over the years (nearly a dozen. I cannot recall one statistics class in high school or college) spend a great deal of time discussing the calculus of statistics and very little who, what, when, where, or why. Accurate formulae and methodology is very important however, what the results are telling you is critical!
Thanks for your comment Donald.

Hopefully the Statistical test advisor covers the who, what, when where and why questions you raise. Knowing when to use a particular statistical test, and especially the assumptions that must be met, is probably the area where most users struggle. As you say, there aren't many books that cover all those points -- David Sheskin's book (we'll post a review shortly) is probably the exception to that rule.

As for interpretation of the results, that's where the advisor will lead to the Analyse-it help. We did try including more help and advice in the advisor itself, but since each page is very much like a PowerPoint slide, space is very limited. Instead we'll us the help to cover concepts such as normality, similar shape distributions, heteroscedascity in more detail, as well as provide guidance on how to interpret each and every statistic Analyse-it shows.
I love the idea of the Test Advisor. There are so many grey areas in choosing tests that going through a step-by-step like this would help, me at least, to narrow the choices by organizing my thoughts. While I do like stats and I anticipate the moment right before hitting the "Enter" key to obtain results or graphs pertaining to data that so much "blood,sweat, and tears" were expended to obtain, truthfully, the time I spend reading and re-reading texts and "decision tress" trying to figure the correct test to use, I could be writing up results, writing grants, or even try to finish my favorite author's newest book! I almost look forward to the day when one can enter the data along with some "parameters" such as data type (nominal or ordinal), what you would like to do (comparisons, correlations, etc.). And while we are dreaming about this, this initial data analysis would even automatically run Normality Tests, data transformations (if needed), and after all that, run parametric or non-parametric tests based on your previously stated goals. Heck, maybe even enter your hypothesis at the beginning and it gets tested!
I know in reality, there are too many variabilities in the data itself and what goals researchers are trying to achieve. Thanks for letting a Grad Student who has been wrestling with these things for the past two years vent.
I do think the Statistical Advisor is a good idea in helping researchers, especially those that don't have a strong stats background, organize their thoughts about statistical procedure.
Thanks for your patience (for the rant) and for a good product!.
Gordon
What a GREAT idea! Like many others, I am rusty on the precise application of the stats tests. This Advisor will be very valuable to me in making sense and of my environmental and ecological data. Any plans to include pointers for multivariate stats or did you already and I missed them?
When I suggested that it be more interactive I meant that I would find it helpful to have more guidelines and decision tables for selecting the appropriate statistical procedure. The wizard was fine. But it asked for information without providing a lot of feedback. Sometimes I found the questions could have been a little more thorough. Maybe having a button to click on for further information (to amplify a specific area that might be unclear to me) might have been nice. Not everyone needs this level of detail. even i don't need it much of the time. But when I do need it, I would like something a bit more thorough than your typical help file, I don't need a wizard version of the help file. I hope that helps.
Hi Tom. We don't currently have any plans to develop Analyse-it, or the advisor, into multivariate statistics (except for what we currently cover -- multiple regression).
Thanks for clarifying your comment Don.

We are planning to provide links from the advisor to more detailed help, to more fully explain each step/question and clarify the terminology used. Where necessary, we'll also explain how you can use statistical tests to, for example, verify normality, examine the shape of the sample distribution, and verify other pre-test assumptions.
This would be a wonderful addition to a great tool set, especially if you extend it to include the Method Validation edition.

Comments are now closed.