5
$\begingroup$

I have been a contract worker for two different companies now, which means I show up at a place for 3-6 months and blow their minds with my awesomeness, but then, to everyone's disappointment, I leave. This is great for me, since I get a new workplace with new problems, but it can be troublesome for the place I left, since they no longer have a person that does something more than look at pivot tables.

In an effort to add more value to the places I work, I have started looking into ways to help business people step beyond just pivot tables. Obviously my goal is to not send my nice coworkers off into the woods without any knowledge, but rather to teach them the basics of regression, some graphics options (really pulls a lot of weight in the business world) and maybe some kind of classification (thoughts on what you think is most relevant would be great) in a weekly/monthly meeting. I am picturing an intro level statistics course for an undergrad, which while long ago for many of us still offers quite a bit of power over something like little green arrows if there was growth at all and little red arrows if there was decline next to an excel spreadsheet.

I am an R person myself, so I have been looking at the different options there, but I am not going to limit myself to just R, so don't let my list limit you. What I have come across so far:

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

I am waiting on IT to give me permission to install this, but it looked promising to me. Excel is the life blood of the places I have worked, so if there is an easy way to get them access to R as an add-in it would help get over the hurdle of learning something new.

http://rattle.togaware.com/

I have played with Rattle some and generally had a good time with it. I think it is a bit beyond what I had in mind in terms of statistical difficulty, but it is put together in a very appealing way.

$\endgroup$
  • $\begingroup$ Whoops, looks like it got too long there, sorry. Let me reformat real quick. $\endgroup$ – asjohnson Feb 28 '12 at 21:07
  • $\begingroup$ The other things I had on my list, before it was so rudely truncated: Excel- Everyone already has it and maybe there are some nice add-ins? It also does quite a bit more than I expect sometimes. There are some reservations about the quality of the analysis in it. Rapid Miner: I interviewed at a place that liked this a lot, but I have not a chance to play with it myself. I am forgetting some of my list, but I think the question is clear, so Cross Validated you have 3-6 months to teach analysts that only use pivot tables some skills a bit beyond that, what tool do you choose? $\endgroup$ – asjohnson Feb 28 '12 at 21:09
  • 3
    $\begingroup$ My main issue would be how you avoid giving them just enough knowledge/tools/status to make them dangerous? You know, the "I got a good R-squared, so this must cause that" kind of thing, which can be misleadingly persuasive to those who don't even know what R-squared means. $\endgroup$ – Wayne Feb 28 '12 at 21:24
  • $\begingroup$ I 100% agree here. It is really my largest concern as well. What comes to my mind is that I think we all agree that statistics done perfectly are better than statistics done less than perfectly, but there has to be a point where the value added from less than perfect statistics is worth it. My goal would be to get them well above that point. Personally, the more statistics I understand the less I believe statistics, so I would hope to foster that in my coworkers. $\endgroup$ – asjohnson Feb 28 '12 at 23:11
  • 1
    $\begingroup$ A couple of older threads that might be of interest here: Software for easy-yet-robust data exploration, What are some valuable Statistical Analysis open source projects?; alas no software will teach them how to use the right model, assess its quality and interpret the results. $\endgroup$ – chl Feb 29 '12 at 13:26
3
$\begingroup$

As I said in my comment, my first concern would be that you need to figure out how to avoid making people more dangerous by handing them tools that are more impressive than Excel but that also require a lot more knowledge/intuition/experience to properly use and interpret. Sort of like replacing company cars with airplanes.

That said, I have always had a bit of a soft spot for gretl. It's command-driven (like SAS, et al, and as opposed to R) rather than a language, has a reasonable GUI and is pretty powerful.

I can see you using graphical flow tools, like Rapid Miner as well.

$\endgroup$
  • $\begingroup$ Thanks for suggesting gretl, I will check it out. I agree on the graphical flow tools option having big benefits, because everyone loves graphics. Once again, I agree that the real battle is going to be preventing bad statistics from getting in, but I do think I can teach someone to fit a pretty simple OLS model, while conveying the complexity that is there. The actual teaching really will be the battle (when is it not?), but plotting that out is another mess, so I think having a good tool is where I will start. $\endgroup$ – asjohnson Feb 28 '12 at 23:15
  • $\begingroup$ Oh, right this is different that stack overflow, well know that I wanted to give you rep points, but that I can't yet! Thanks for the well thought out response. $\endgroup$ – asjohnson Feb 28 '12 at 23:18
  • $\begingroup$ @asjohnson: gretl uses gnuplot to do its graphics, so if you're willing to dig deep (and learn gnuplot), you can make any kind of graphic you can imagine. Though that could also be a weakness of the gretl option. $\endgroup$ – Wayne Feb 29 '12 at 0:47
  • $\begingroup$ @asjohnson That's not a limitation of Cross Validated, but I believe you didn't reach the min reputation level (15) to vote up at the time you wrote the above comment. $\endgroup$ – chl Feb 29 '12 at 13:12
  • $\begingroup$ @chl Oh I know, I just always forget that a stack exchange is different from stack overflow (I have like 50+ rep there, huge deal, I know). Do you know if each stack exchange has different rep? I guess I just need to get more active on the ones I read. $\endgroup$ – asjohnson Feb 29 '12 at 15:26
4
$\begingroup$

Judging from my experience with research-active clinicians, imparting safe levels of statistical knowledge is very, very difficult. In 3-6 months while doing other things I would hazard that it's impossible.

I think in your shoes I would just make a lot of reports that write themselves (with Brew, Sweave, R2HTML, whatever) and go HEAVY on the visualisations.

Anything else and you're just putting a loaded idiot-gun in their hands.

I would also wean them off Excel. Just get them using .csv files, in Excel if they wish. Proprietary file formats, merged cells, coloured cells, they're all the devil's work. A nice, flat, colourless, .csv, that's all they need.

$\endgroup$
  • 1
    $\begingroup$ That is a pretty legitimate point. One of my professors where I did grad school has a guide on sweave I was reading and it really does seem like the best thing. It would likely be a good idea to use this type of reporting as an excuse to get good with sweave. Do you think that is the best one to use or have experience with others? That is the only one I know anything about. $\endgroup$ – asjohnson Feb 29 '12 at 15:29
  • 1
    $\begingroup$ As far as convincing someone in the business world to not use .xls/.xlsx, I think I would have an easier time convincing someone that I am 10 feet tall (note: I am not 10 feet tall and it is pretty noticeable) than I would have trying to wean someone off .xls/.xlsx. The best thing I do there is write .net to automate stuff as much as possible. Maybe google will have enough of a wrecking ball to get people off excel, but then I imagine you just jump over to their proprietary file format. $\endgroup$ – asjohnson Feb 29 '12 at 15:30
  • $\begingroup$ Sweave is genius. You couldn't take it off me with an army. It will pay back any investment you put in, with interest $\endgroup$ – Chris Beeley Feb 29 '12 at 19:47
  • $\begingroup$ Alright, chalk another point up for me getting my act together and sweaving everything. $\endgroup$ – asjohnson Feb 29 '12 at 19:50
1
$\begingroup$

When it comes to data mining, Weka is very user-friendly.

$\endgroup$
  • $\begingroup$ What are your thoughts on Weka vs. Rapid miner or even Rattle? I have yet to really get involved with any of them, since I have been doing R for 7 years and just in the last few years found out that there are these other great tools (turns out my undergrad and grad schools just liked using R a lot). $\endgroup$ – asjohnson Feb 29 '12 at 15:32
0
$\begingroup$

Sofa is worth a look http://www.sofastatistics.com/home.php

$\endgroup$

We're looking for long answers that provide some explanation and context. Don't just give a one-line answer; explain why your answer is right, ideally with citations. Answers that don't include explanations may be removed.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.