(This article was originally published at Statistical Modeling, Causal Inference, and Social Science, and syndicated at StatsBlogs.)
Bob Nau writes:
I know you are not particularly fond of Excel, but you might (I hope) be interested in a free Excel add-in for multivariate data analysis and linear regression that I am distributing here: http://regressit.com. I originally developed it for teaching an advanced MBA elective course on regression and time series analysis at Duke University, but it is intended for teaching data analysis at any level where students are familiar with Excel (and use it on PC’s), and it is also intended for serious applied work as a complement to other analytical software. It has been available to the public since May 2014, and a new version has just been released. If I do say so myself, its default regression output is more thoughtfully designed and includes higher quality graphics than what is provided by the best-known statistical programming languages or by commercial Excel add-ins such as Analyse-it, XLstat, or StatTools. It also has a number of unique features that are designed to facilitate data exploration and model testing and to support a disciplined and well-documented approach to analysis, with an emphasis on data visualization. My frustration with the stone-age graphics output of the leading regression software was the original motivation for its development, and I am now offering it for free as a public service. Please take it for a test drive and see for yourself. I’d welcome your feedback.
I don’t know Excel at all so I can’t take it for a test drive . . . but I bet that some of you can! Please share your thoughts. So many people use Excel that an improvement here could have a huge effect on good statistical practice. I don’t know if Reinhart and Rogoff read this blog but there must be some Excel users in the audience, right?
P.S. Nau wanted to share some further thoughts:
It may appear at first glance as though there is little that is new here: just another program that performs descriptive data analysis and plain old linear regression. The difference is in the details, and the details are many. Every design element in RegressIt has been chosen with a view toward helping the user to work efficiently and competently, to interactively share the results of the analysis with others, to enjoy the process, and to leave behind a clear trail of breadcrumbs. In this respect, RegressIt is a sort of “concept car” that illustrates features which would be nice to have in other analytical procedures besides regression if the software was designed from the ground up with the user in mind and did not carry a burden of backward compatibility with the way it looked a decade or two ago. Also, it tries to take advantage of things that Excel is good for while compensating for its lack of discipline. The design choices are based on my own experience in 30+ years of teaching as well as playing around with data for my own purposes. When a student or colleague or someone on the other side of the internet wants to discuss the results of an analysis that he or she has performed, which might or might not be for a problem whose solution I already know, I want to be able, with a few mouse clicks, to replicate their analysis and drill deeper or perform variations on it, and compare new results side-by-side with old ones, while having an armchair conversation. I might also want to do this on the spur of the moment in front of a class without worrying about my typing. When I am looking at at one among many tables or charts, I often wonder: what model produced this, and what were the variables, what was the sample, when did the analysis take place, and by whom? What other models were tried before or afterward, and what was good or bad about this one? If a chart is just labeled “Residuals of Y” or “Residuals vs. Fitted Values”, that is not very helpful, particularly if it has been copied and pasted into a report where it takes on a life of its own. And when I look at the output of a model on the computer screen, I want to see as much of it at one time as possible. I want an efficient screen design—ideally one that would look good in an auditorium as well as on my desktop—and I want easy navigation within and across models. I would rather not scroll up and down through a linear log file that reminds me of line-printer days (which I do remember!) and makes it hard to distinguish the code from the results. I would like to see a presentation that by default is fairly complete in terms of including some well-chosen chart output that allows me to engage my visual cortex without saying “yuck”. And I want the same things if the original analyst is not a student or colleague but merely myself yesterday or last week or last year.
I hope you will give it a close look, kick the tires, and take it for a drive with some data of your own. And please read everything that is on the features and advice pages on the web site. Otherwise you may overlook some of what RegressIt is doing that is novel. And whatever you may think of it in the end, I would welcome your input on improvements or extensions that could be made. Is there any low-hanging fruit could easily be added, or is there some deal-breaking omission that absolutely needs to be fixed? We can make changes in a hurry if we have to–there is no calendar of scheduled releases. We are two professors who work on this in our spare time. RegressIt’s feature set is limited at present, but our hope is that the features it does include will be useful in some circumstances to people who do most of their work in R or Stata and well as to people who do most of their work in Excel, and we plan to add more to it in the future. Thanks in advance for your input!
The post An Excel add-in for regression analysis appeared first on Statistical Modeling, Causal Inference, and Social Science.
Please comment on the article here: Statistical Modeling, Causal Inference, and Social Science
