My friend and I probably see each other’s statistical product choice as a series of stereotypes. For him, R’s open source development model presents serious quality risks in contrast to the tried-and-true proprietary SAS QA methods. Compared to SAS’s serious enterprise business focus that includes data integration and BI, he argues that R’s simply for academics and researchers. And, of course, R’s in-memory processing limitation consigns it to toy data only.
I, on the other hand, paint SAS as the slow-moving gorilla, lagging behind R in statistical innovation, driven by a tired 1980s language portfolio of data step, procs and macros that pales in comparison to R’s modern array and object orientation. And I note with joy that R now wears the crown of preferred platform for statistics graduate students that SAS wore in the 80s and 90s.
Nursing Hefeweizens, my friend and I took turns sharing our latest statistical challenges. He noted that his current work revolves as much on graphics and visualization as it does on statistical models per se. I countered that I’ve been heavy into time series and forecasting over the last nine months.
After a few pints, I just had to tweak my friend on SAS’s inferior graphics, noting that while visualization is central to the R analytical approach, it often seems an afterthought in SAS. And with good reason, I argued, for SAS graphics are weak. Back in my SAS days, I often dropped into companion product JMP for visualization. Now even SAS devotees often prefer R’s superior graphics.
A bit annoyed, my friend asked last time I’d seen SAS graphics. I was embarrassed to acknowledge it’d been 12 years. He challenged me to review the latest offering before continuing to pass judgment. Touche.
As I started to discuss my forecasting work, my friend asked what R packages I was using. I responded that I’m working with the forecast library written by Rob Hyndman that I absolutely love. Among its many features, forecast estimates exponential smoothing models as well as the autoregressive, integrated moving averages (ARIMA) popularized by Box and Jenkins. What sets forecast apart, I noted, is its ability to automate the model selection process based on established optimization criteria. Since I’m often estimating scores or even hundreds of models simultaneously, this feature is particularly handy.
Once I’d finished discussing my work, my friend jokingly asked about the likely now-defunct economics grad student who’d authored forecast, and how I could have any confidence in the model estimates the package provided. Would the author fix bugs found by the community and support the library in the long-run, he mused? And would I bet a week’s compensation on forecast’s quality?
As we wrapped up our enjoyable banter, I think both of us realized we’d built straw men to support our statistical platform arguments. Indeed, as I later thought about the discussion, I realized just how out of touch I was with SAS’s current capabilities, and how in the dark he was in general about R.
At OpenBI, we constantly remind ourselves not to anchor competitors in the past. What a product or firm looked like in 2007 could well be very different than it is in 2012. Losers can become winners and winners can become losers very quickly in technology. It’s critical, therefore, to stay up on the market.
So I set out to take a look at some of the latest graphics capabilities of SAS. I wish I could have gotten a demo version of the software to test-drive, but had to settle instead for a review of the samples output gallery. I was more impressed than I expected to be, even if the sight of macro code behind the visuals gave me heartburn. I think I could do in SAS/GRAPH much of what I now accomplish with the excellent R lattice and ggplot packages. I was especially intrigued by the newest Statistical Graphics, with the sgpanel proc for conditioned or trellis visuals. Maybe SAS graphics is not the beastly laggard I’ve characterized.
I received an email from my SAS friend a few days ago. Just as I’d taken a look at current SAS graphics, he’d started investigating time series and forecasting for future work. Low and behold, he came across an excellent book on the latest state space methods written by academic Rob Hyndman and others. Hmm, Rob Hyndman – isn’t he the same guy who authored the R forecast package? Perhaps the economics grad student isn’t defunct after all.