By the way, I use R almost exclusively for my statistical research in my day job as a professor of statistics at Texas A&M University.

We have the same definition of social data analysis by the way. In some sense, folks in statistical education have been doing it for a number of years in the form of small group projects. The internet has brought the ability to do this with with a much larger number of people. I am most excited about the possibilities for scientific insight and learning that these online collaborations may be bring about.

What I described above are some of capabilities that StatCrunch brings to the table for social data analysis in an educational setting. This aspect of the site has only been available since August 2007 and already thousands of data sets and results are being shared and discussed on the StatCrunch site. You can access them by clicking the appropriate links at the top of the StatCrunch web page. So, I definitely have in mind to take StatCrunch more into the realm of social data analysis.

I’ll start with my “fixation on R.” I wasn’t trying to suggest that people use R in lieu of something GUI-based. All I meant was that it felt weird to me, because I’ve always learned analysis with R at both my undergrad at Berkeley and graduate education at UCLA. Nothing more.

In fact, R isn’t even my weapon of choice (although I do still think it is very useful). My focus is in design and data visualization so I’m usually in Illustrator or immersed in actionscript. I won’t try to argue that R belongs in stat education (I think it does); that discussion could go on forever.

From a design perspective, the StatCrunch UI does need work; however, I guess I should note that StatCrunch is a bit better than Covariable in this regard. But still, ask any designer, and I’m absolutely sure they will agree that there’s a lot of room to improve ease of use. As for that 1960s remark, well, R is a programming language. Write a script and run it. What’s current? Look at any web app on TechCrunch. There is a clear difference between apps today and StatCrunch’s collection of applets.

Onto the definition of social data analysis. We clearly have very different views on what this is. I was referring to the ideas made popular by Wattenberg et. al. – a group of people analyzes a single dataset together and comes to some kind of consensus. StatCrunch isn’t trying to do this, is it? Covariable I think is trying.

Anyways, in my efforts to keep things short — as an educational tool, StatCrunch does have potential but can still stand to undergo a lot of changes. Covariable is not trying to cater to education, which I believe is a problem.

Looking forward to the August 2008 (I’m assuming that’s what you meant) update.

I find the remarks about the StatCrunch user interface extremely odd. This aspect is what most of the 100,000 annual users seem to like best about StatCrunch compared to other packages. StatCrunch is definitely still under development with a major release in August of 2007. It seems that Nathan did not do his homework on this piece of the evaluation. His comments are even more odd when you take into account that he uses R which has a 1960s interface. What does a 2000ish program look like?

The question posed in the last paragraph shows that Nathan really has no idea what he is talking about. The advantages over emailing an Excel file are numerous.

First, the Excel issue. Have you ever tried doing basic groupings in Excel (a pivot table nightmare!), and of course, often times P-values in Excel are wrong. StatCrunch also has interactive graphics which means you can highlight observations in a graph to see where they are located in the data table and in other graphics. This feature is very nice to have when teaching/learning statistics. Excel does not have this sort of feature. Simply put Excel is a spreadsheet program that some folks try to use for statistics but it simply does not work very well.

Now on to social data analysis where Nathan’s comments are most disturbing. Think for a moment about a statistics course where an instructor wants to share data sets with a large group of students. With traditional desktop software, each student would need to obtain the software and install it on their machine. Then using Nathan’s suggested approach, the instructor would need to email each data set to every student. Each student would then in turn have to go though the steps to load the data set. With StatCrunch, an instructor can simply point the students to his or her public profile. Students then can immediately start analyzing data by clicking on linked data sets with no need to install any software at all. Students can also access the software from any online location and are not limited in terms of where they can work. They save their work on the StatCunch server so they can access it in the future or share it with others. With StatCrunch, they can even prepare online project reports combining data and analysis results to be shared with their fellow classmates and their instructor.

I apologize for my lengthy response here, but this review is really really lacking.

I find the remarks about the StatCrunch user interface extremely odd. This aspect is what most of the 100,000 annual users seem to like best about StatCrunch compared to other packages. StatCrunch is definitely still under development with a major release in August of 2007. It seems that Nathan did not do his homework on this piece of the evaluation. His comments are even more odd when you take into account that he uses R which has a 1960s interface. What does a 2000ish program look like?

The question posed in the last paragraph shows that Nathan really has no idea what he is talking about. The advantages over emailing an Excel file are numerous.

First, the Excel issue. Have you ever tried doing basic groupings in Excel (a pivot table nightmare!), and of course, often times P-values in Excel are wrong. StatCrunch also has interactive graphics which means you can highlight observations in a graph to see where they are located in the data table and in other graphics. This feature is very nice to have when teaching/learning statistics. Excel does not have this sort of feature. Simply put Excel is a spreadsheet program that some folks try to use for statistics but it simply does not work very well.

Now on to social data analysis where Nathan’s comments are most disturbing. Think for a moment about a statistics course where an instructor wants to share data sets with a large group of students. With traditional desktop software, each student would need to obtain the software and install it on their machine. Then using Nathan’s suggested approach, the instructor would need to email each data set to every student. Each student would then in turn have to go though the steps to load the data set. With StatCrunch, an instructor can simply point the students to his or her public profile. Students then can immediately start analyzing data by clicking on linked data sets with no need to install any software at all. Students can also access the software from any online location and are not limited in terms of where they can work. They save their work on the StatCunch server so they can access it in the future or share it with others. With StatCrunch, they can even prepare online project reports combining data and analysis results to be shared with their fellow classmates and their instructor.

I apologize for my lengthy response here, but this review is really really lacking.

I fail to see the need for applications such as Covariable (web based or otherwise). I can’t imagine it has anywhere near the computing power of software like R. At $19.95 a year (that is if you want to do such frivolous things as ANOVA, multiple or non-linear regression, and hypothesis testing) vs. $0 for R, the makers of Covariable are kidding themselves.

StatCrunch, on the other hand, does seem to have value as an educational tool. It looks like it might be fun to play with.

When you need to unscrew something, you won’t go through the whole toolbox before finding the screwdriver or reach for the saw when you need to hammer a nail.

I’ve found it’s sometimes easy to get ‘carried away’ by the ease of doing lots of exploring, which can lead to over fitting and meaningless discovery. I was thinking a more integrated technique of keeping track of my hypothesis in advance, might help.

What do you use?

Covariable’s Visible Statistics has a boxplot! Too bad it can’t be included in the report. Advantage squandered.

However, the Visible Probability pages are very nice demonstrations of the fundamental PDFs, PMFs, and CDFs. I’m using these as a resource for my undergraduate courses.