[R] Entire Organization Switching from SAS to R - Any experience?

[Marc Schwartz]

On Jul 17, 2009, at 9:57 AM, Kelvin Lam wrote:

>
> I should elaborate the situation a bit more.  We store our data in  
> UNIX and
> have been using UNIX SAS for our work.  My Biostat dept has 40 SAS  
> users
> from which at most 10 also use R.  The Epi/Grad Students/Investigators
> combine for another 30-40 not-so-frequent SAS users let alone R.  So  
> we are
> talking about 80 folks/workhorses in the entire institute.
>
> One of my thoughts is to break up the Biostat group into two so that  
> one
> uses R solely to reduce the number of licences.   IMHO, the pro is  
> to worry
> a smaller group of users.  However, the cons will be who to be  
> assigned to
> respective group.

You have several business, human behavior and budgetary issues to  
consider. It is not usually just a matter of making the business case  
that saving annual license costs is the sole factor in making a  
decision to switch to R.

You have to consider the receptivity of the existing base of SAS users  
to changing to R. Are they open to it or are they not motivated to  
change? In the latter case, is the organization in a position of  
forcing change or not? Being a non-profit organization, you can simply  
say, due to funding issues, we are going to have to eliminate some 'x'  
number of SAS licenses to stay within our operating budget. That will  
have an impact on how many users can in fact continue to use SAS.

However, if you compel wholesale change, are you at risk of losing  
people who are resistant and decide to move on? Are they key people  
where their loss would have substantive impact on the organization and  
project commitments, at least in the short term? Yes, on one level, we  
can all be replaced by somebody else, but at what short term cost to  
the organization and it's customers?

If there is resistance amongst some proportion of the staff, an  
incremental approach would be very appropriate. Balance your funding  
issues with the number of staff that would be impacted in the near  
term. Can you eliminate 'x' SAS licenses this year, 'y' more next year  
and so on so that the transition is implemented over a multi year time  
frame while still working within your funding constraints?

Solicit feedback from the staff to see who is open to using R and who  
is not. Let that be a key factor in any decisions to partition the  
staff. Get an idea as to the scale of the battle that you are facing  
with respect to change. Identify the "low hanging fruit" to look for  
incremental and consistent wins that you can build on. Those who are  
resistant to R may simply need time to see that what they have done in  
SAS can indeed be done in R with greater quality, speed, flexibility  
and in time, at a lower cost. Once they get over that hurdle, they may  
come on board with you and make subsequent transitions easier.

With an eye towards the future, be sure that new hires are skilled in  
R, so that as you may need to deal with staff turnover or growth, you  
are enabling the future use of R by a growing number of folks who have  
pre-existing R skills. Set yourself up for future success.

Consider R related training and the costs associated with it. The  
costs are not just what you may have to pay for training, but the  
opportunity costs in the short term of getting people up to speed and  
the loss of productivity short term, even though as Frank noted, you  
will realize notable gains in the long term. Consider how your  
existing project commitments would be impacted and how you may have to  
allocate or re-allocate the workload during the transition.

Consider the costs and timelines associated with converting an  
existing base of SAS code that has perhaps gone through a review and  
validation process. What will it take to replicate that functionality  
in R with the same level of reliability? What methodological issues  
will you face in the transition from SAS to R, given the differing  
philosophies? How long will it take, who will do it and what other  
tasks or projects may be impacted during the transition?

In some environments (eg. Big Pharma), re-training and especially code  
conversion/validation costs alone outweigh the savings of not paying  
for SAS licenses. This is why there is a significant hurdle to using R  
in that environment even though such companies may pay SAS millions of  
dollars per year.

I don't know that there is a one size fits all approach to moving from  
SAS to R. Each operating environment has its own characteristics  
relative to budgets, politics, people and so forth. The points that I  
raise above may be typical but only some may apply to your situation  
and there may be others that I have not raised.

HTH,

Marc Schwartz