• If you are citizen of an European Union member nation, you may not use this service unless you are at least 16 years old.



Page history last edited by Nancy Reid 12 years, 4 months ago


Statistical Sciences GSC 14 restructuring



NSERC's decision on the conference grouping for statistical sciences

May 12, 2009

We have received a reply from NSERC to our proposal for restructuring what used to be GSC 14.  They have decided

that next year we will be in a Group with Mathematics.  conference-decision.pdf

The boxes for the group are in two scanned documents (two sides of a single page). boxes-stat.pdf      boxes-math.pdf



Reply from Rawni Sharp at NSERC, September, 2008

 Reply to Nancy re new structure and boxes.doc


Proposal sent to NSERC, July, 2008


The sidebar has the version of the proposal that was sent to NSERC, along with the latest version of the building blocks.  Please feel free to comment on these documents.


Thanks to all for your input and interest.








New documents uploaded on July 8, please have a look

Also there are several comments posted on these new documents, but they are (unfortunately) all the way

at the bottom of the comments list.


We have prepared a proposal for a separate broad area panel, or Group, called Statistical and Quantitative Interdisciplinary Sciences.  We have uploaded a draft of the proposal, and a draft set of units for this group.  We are planning to send this to NSERC before July 18.  Many refinements are needed, we welcome your input.  Please send comments by email or post comments to this page as soon as possible.



Summary of meeting with NSERC, June 26  Please visit this page to find out new information from NSERC about the proposed restructuring.


The current proposals from NSERC are available on the side bar, and also available at Nancy Reid's

web page www.utstat.utoronto.ca/reid/.  Please have a look.


There are many comments below, most sent by email to Nancy Reid.  These are very helpful,

and thanks to all who responded.  I am happy to add anyone to the list of people able to write

on these pages, please let me know if you would like to be added.  The Liaison committee is working on creating a set of building blocks that cover the areas adjudicated by GSC 14, and reflect the research areas covered by GSC 14.


The current proposed building blocks formed from GSC 14 are listed below.  These are quite provisional, and many people have pointed out both obvious and not so obvious inconsistencies.  Further comments are very welcome any time; the final configuration of building blocks is very much up to us.


Theoretical probability 19a

Applied probability 19b

Actuarial science and math finance 20

Statistics of biological systems 18a

Applied statistics 18b

Statistical methodology 21

Psychometrics    32

Statistical analysis 33

Simulation and computational statistics 16b

Comments (Show all 41)

Robert Platt said

at 5:13 pm on Jun 15, 2008

Some somewhat scattered comments:

1. Where is "statistical theory" or "mathematical statistics"? I don't think this line of research is covered well anywhere.
2. In general, there seems to be an emphasis on computational/systems biology statistics at the expense of medical/clinical biostatistics.
3. Further, traditional medical/clinical biostatistics is spread across several building blocks. LS17 "Quantitative Biology" and MSCS18-a "Statistics of Biological Systems" both refer to it, but it appears to be secondary to the statistics of systems biology in both groups. MSCS21 "Statistical Methodology" and MSCS33 "Statistical Analysis" both list methods relevant to biostatistics, but in large grab-bags of other stuff. It seems like researchers in biostatistics will fall between many of these blocks and not directly in any one of them.
4. I agree re psychometrics. I don't see that we have that much research that is specific to psychometrics as opposed to the other topics Nancy mentions, or biostatistics for that matter.
5. In principle, I'm not in favor of the broad math sciences/computer sciences grouping. As many have already noted, GSC14 was already a wide-ranging group from probability to very applied statistics. To add this to the much larger mathematical community will probably not do well for statistics.

Nancy Reid said

at 6:57 pm on Jun 15, 2008

Email to NR from Martin Barlow:

Dear Nancy,

I have the following comments:

1. In general terms, its not clear how well it will work.

2. Conditional on it working alright in general, I have the
following specific comments on the areas I know best:

(2a) 'Theoretical probability' seems a reasonable area, and the new
situation could be an improvement over the current one, where probabilists are divided between GSCs 14, 336, and 337.

The current 19a also includes what I would call 'Theoretical/
Mathematical statistics' - eg. nonparametric models, ... It might be better for 'Math. Stat.' to be moved elsewhere.

(2b) Putting Math. Finance and Act. Science together seems
reasonable to me, though I am not an expert in either area.

Although these represent to some extent different traditions,
and the people doing these may be different sociologically, both groups are addressing the same general problem (risk control).


Nancy Reid said

at 10:26 am on Jun 16, 2008

David Binder wrote in email to NR:

These comments are based on the following:

a. I heard the NSERC briefing at the SSC Board Meeting in Ottawa.
b. I am not a beneficiary of any NSERC funding.
c. I am deeply concerned about the future of research in the Statistical Sciences in Canada.

1. One of the reasons for the restructuring seem to be to allow more flexibility of allocation of funds across GSC disciplines, without being constrained by a fixed budget within GSC's.
2. The objectives for NSERC funding of Statistical Sciences research in Canada should be to:
i) provide sufficient funding to newer researchers to allow them to interact with more senior researchers, to conduct research and to begin training for postgraduate student in the field. It is my impression that the teaching load for younger researchers can be affected by the existence of an NSERC grant.
ii) to further research in statistical sciences to benefit Canadians, through the solving problems and through the development of methods that can be applied other disciplines or outside of universities in industry, government and institutions (particularly health institutions).

Based on this, I feel that the Broad Area Panel has too much focus on problem-solving and the mathematical aspects of our discipline. As a result, funding for newer researchers and for research with anticipated impact in non-mathematical disciplines may be suffer, and yet it is these are aspects that will be of most benefit to Canadian society. I do not know which BAP is better for our discipline, but one that places more emphasis on the beneficiaries of our research outside of our discipline would make more sense.

David Binder
2005-2006 SSC President

Nancy Reid said

at 10:28 am on Jun 16, 2008

Keith Knight in email to NR wrote:


A few first impressions:

1. The list of building blocks seems haphazard - e.g. why have both
applied statistics and statistical analysis? Also where does theoretical
statistics fit in? Why have psychometrics and not econometrics?

2. In terms of research culture, we're not as close to computer science
as we might like to think we are. CS places very high value on having tons of conference
proceedings. That said, a broad area panel with CS, OR and stats might
make sense since there are natural links between applied prob, actsci/math
finance and OR.

Nancy Reid said

at 10:33 am on Jun 16, 2008

Sudhir Paul, in email to NR

As other SSC members I am re concerned with the current NSERC restructuring. Is it possible to let us know the rationale of prefering to be with Computer Science rather than Mathematics.

Nancy Reid said

at 10:36 am on Jun 16, 2008

In reply to Sudhir's comment: the rationale for going with CS and OR is, I think, that
(1) the disciplines are closer to stats than math is
(2) a desire to escape from math
(3) a feeling that we'd be better off financially

However, we need to hear from the community; our liaison committee has discussed this and we are not at all clear if
some or all of statistics should be with CS

My own preference is for statistics to stay as its own Broad Area Panel, but I don't think this is an option. However our
committee can take this up with NSERC to see if the chances are greater than zero.

Nancy Reid said

at 10:39 am on Jun 16, 2008

David Andrews wrote, in email to NR:

Decisions about the support of science should be based
in part on evidence. The grouping Statistics with Mathematics
makes no scientific sense and flies in the face of all
recent evidence.

The great contributions of statistics in this age arise
in connection with work in scientific disciplines
rather than in mathematics. Mathematics has important
but different priorities.

Statistics is playing an ever increasing role in
medicine, public health, education. The training
of students with interests in these areas is
singularly important. The development of research
and training with these goals has been the priority
of Statistics, and other disciplines to a far
greater extent than it has been in Mathematics.

Statistics has flourished in Canada in those departments
that have broken from the yoke of Mathematics to form
collaborations and associations with other disciplines.
Canadian research is world renown. I speak of the
seminal work on survival analysis spawned at Waterloo,
the explosion of research in genetics at Toronto,
the contributions to environmental research at UBC
as examples.

In the United States, the very best departments
Stanford, for example are best known for their close
interaction with many departments -- but not with

It makes no scientific sense. It would not be fair to Mathematics.

Nancy Reid said

at 11:28 am on Jun 16, 2008

Jerry Lawless wrote, in email to NR:

1. the form of the panel approach: ... Assuming it is going to happen then we need to ask questions about the structure;

Questions that come to mind are
can stat science DG applications go to more than one panel?
how much will panels overlap or consult each other?
what will ensure that there is really someone on the panel who knows much
about a proposed topic? (I observed when I was Gp Chair that the Math A and B
GSCs had no one who knew about certain proposed topics, so they went on the
reviews plus "general reputation". But now there is supposed to be less
weight on reputation or track record and more on the proposal itself.)
how will the review process maintain the very good features of the DG program
(support for excellent researchers, e.g.), which the Int. Review Ctee noted?

2. whom Stat Sci is grouped with. We should continue to argue against being with
Math in the strongest terms.
NSERC should get themselves into the 21st century. The list of building blocks
which they have is grossly inadequate. We need to
propose a list, and we need to be very thoughtful about stressing the many
areas that stat scientists work in, while still keeping the number of blocks
reasonable. For example, do we want psychometrics, econometrics,
envirometrics etc? If so why not also social statistics,
industrial statistics, statistical
genetics, genetic epidemiology, etc? NSERC may argue that these areas are
more related to CIHR or SSHRC but that is something that must be sorted out.
Many of the strongest DG recipients are highly
involved in applied areas, and surely we cannot recognize only those whose
interests lie in other "NSERC" fields.

Nancy Reid said

at 11:30 am on Jun 16, 2008

Some comments on Jerry Lawless's questions:

1. Yes, the panel approach is going to happen, and we will take these questions to NSERC.

2. The list of building blocks is indeed wrong at the moment, but I am fairly confident that NSERC will re-organize these along lines we suggest. This is very important, and the Liaison committee can try to do its best, but all suggestions are very welcome. Martin Barlow has already noted above that theoretical statistics somehow got in the theoretical probability block, which needs to be amended.

Nancy Reid said

at 3:15 pm on Jun 16, 2008

From Will Welch to NR by email:

Dear Nancy,

I agree wholeheartedly with the alignment with CS rather than Math. At UBC there
is currently zero research interaction with the Dept of Mathematics. As you know
Statistics broke off from Math so all the more surprising. In contrast, we have
strong interactions with CS via two joint appointments and other research collaborations.
Moreover, there is a segment of the CS community that really values statistical
methodology, in the sense that it is useful for their *own* research.

All this my personal opinion ...


- Will

Nancy Reid said

at 5:15 pm on Jun 16, 2008

From Jinko Graham to NR by email:

To me, Statistics is a relatively young discipline
with a fundamentally applied set of priorities.
Mathematics is much older, more established
and has important but different priorities.
Would this be a happy relationship of equals?
I agree that a BAP that places more emphasis on the
beneficiaries of our research outside of our discipline
seems sensible.

I share the concerns of others on the Wiki about the list of
building blocks that NSERC has provided. They seem to have
emphasized some statistical areas and either completely left
off or not given serious consideration to others.

Jinko Graham

Nancy Reid said

at 11:21 am on Jun 17, 2008

From Mary Thompson to NR by email:

Dear Nancy,

(These comments could be added to the Wiki.)

I do not fully understand how the restructured system is to work.

One of the goals is to make possible better review of interdisciplinary proposals, and I guess part of the idea might be that if building blocks are small they can be juxtaposed in different ways for at least part of the time, though the logistics look very difficult.

On the other hand this new system might make handling emerging disciplines -- especially those that are emerging out of interdisciplinary collections -- easier. Actuarial science/finance/econometrics do form a sort of interdisciplinary collection with their own block (in the example), which might thereby gain disciplinary strength.

Regarding the blocks:
-- presumably each block should account for a substantial portion of current proposals
-- presumably each block should be such that a small group of people could have the expertise to judge the proposals in it
-- at the same time it looks as though we have here an opportunity to shape future directions by defining the blocks; I think this is the MOST IMPORTANT PART of what faces us -- and might make it desirable to try to relax the first two conditions a little
-- having probability theory as a block makes sense, and I agree that this is likely preferable to the current spread over three GSCs
-- I agree biostatistics should be more prominent, and of course it really is a substantial part of what the GSC has been considering

Mary Thompson

Nancy Reid said

at 11:23 am on Jun 17, 2008

From Andrei Volodin to NR by email:

Dear Nancy:

Completely support your suggestion that we should be assigned not to a broad area panel called "Mathematics and Statistics", but to a broad area panel with Computer Science and Operational Research.

Next, I do not like the title of block number 33. Now it is Statistical Analysis. Could we change it to Mathematical Statistics?

Are there any instruction how to proceed, if a research falls into a few blocks categories? Mine research is in 33, 19A, and 16B, and I do not know which one is more important.

Have a wonderful day!

With kind regards, Andrei
Dr. Andrei Volodin
Associate Professor of Statistics
Department of Mathematics and Statistics
University of Regina
Regina, Saskatchewan
S4S 0A2 Canada

Office: CW 307.27
email: andrei@math.uregina.ca
Phone: (1)(306) -585 4771
FAX: (1)(306) -585 4020
Cell: (1)(306) -501 5094

Nancy Reid said

at 11:23 am on Jun 17, 2008

From David Andrews to NR by email:

I believe that science would be better served if decisions that affect
future research in statistics were made in consultation with Computer
Science rather than Mathematics.

I am concerned that the such major decisions are being conducted
in such haste.

I encourage you to consider the following.

One question that should be faced, although addressing it broadly
may not be productive at this time, is whether Canadian science
would be better off if all the mathematical disciplines were considered
together. If decisions were made by one group including Computer
Science, Mathematics and Statistics, with no group in a majority,
the decisions would be different, and perhaps even better.

Nancy Reid said

at 5:24 pm on Jun 17, 2008

From Neal Madras to NR by email:

Hi Nancy,

I am copying a few probabilists from the list as well.
My message is about the details of the "building blocks",
especially those relating to probability. I am not sending
this to NSERC yet; I will wait to see if there is feedback from

Re MSCS19B "Applied Probability": It is often pretty hard to tell
theoretical probability from applied probability (e.g. I think much
of Annals of Applied Probability is theoretical; sometimes I use
the rule of thumb that a math paper is "applied" if its most important
contribution is not a proof). Anyway, I will go with the spirit
of the proposal for now. I suggest the following:

- Move "interacting random processes" from "Applied probability" to
"Theoretical probability". I am thinking of work by people like
Liggett, Durrett, Dawson, Quastel.... Most of this work
is very theoretical, even though its motivation may come
from statistical mechanics and/or biology.

- Move "stochastic modelling" from Theoretical probability to
Applied probability: Modelling is the essence of applied math,
isn't it? A good modelling paper may advance science much more
than it advances theoretical math.

- Maybe add "Reliability" to "Applied Probability"

- Applied Probability must be linked with the Operations Research

- Why are "Nonparametric models" and "density estimation" under
Theoretical probability rather than Statistical methodology? I guess
I don't mind, but it seems odd. Also, what are "function inverse
problems" (at the bottom of the Theoretical Probability list)?

- Perhaps MCMC should move from Statistical Methodology to
Simulation and Computational Statistics (but I leave that to the
statisticians to decide).

As for the overall proposal: it is somewhat intimidating to try
to predict how this all will work and to provide analysis now.
At the moment, I am willing to go along with it.


hchipman said

at 9:09 pm on Jun 17, 2008

Some minor suggestions for changes to the building blocks and also links with other disciplines:

Simulation and computational statistics:
* drop "Scientific computing" (not sure what this means in a statistical context),
* add "resampling methods (MCMC, Bootstrap, ...) ... although these are difficult topics to place since results are a combination of stat computing and theory that has little to do with computation.
* add "data visualization/statistical graphics"
* add "statistical programming environments" or "statistical software"

I think that while psychometrics is a identifiable area (we can get 5 applications a year) it might not be large enough to be an individual panel - merge /w applied statistics?

The distinction between statistical methodology and statistical analysis seems artificial. Statistical methodology appears to be a grab-bag covering a significant fraction of what is now GSC 14 (a quarter? a third?)

Add connections with MS16B (Simulation and computational statistics):

* Weak link from CS22c (data management) because of "data mining"
* Weak link from CS17b-10 (Algorithms and data structures) because of "randomized algorithms"
* Weak link from CS29-31 (AI) to MS16b because of "machine learning"
* Weak link from CS30-10 (computer graphics & visualization) because of "data visualization"

Nancy Reid said

at 2:48 pm on Jun 18, 2008

The documents on the side bar show the current allocation. Note that CS and OR has something like 27 building blocks; whereas in the Math and Stats grouping there are 14 in math, 2 in probability, 1 in actsci/finance, and 6 in statistics. It's not at all clear what the implications are for number of building blocks, size of applications and so on, but this might affect how we think about the best place for us...

Nancy Reid said

at 2:58 pm on Jun 18, 2008

Suggestions on building blocks from Mary Thompson:

Maybe a couple of people who have been on the GSC recently and have a sense of volume of applications in various sub-fields of statistics could suggest a structure. The NICDS areas plus mathematical statistics might provide nuclei, which would group researchers roughly by area of application but still essentially be a grouping of statistical science.

The block I am most doubtful about is applied probability. It is in a way part of everything we do, yet applications which are just about applied probability are few, and very different in subject matter. I'd be inclined to try to divide that one up among others.


Nancy Reid said

at 4:09 pm on Jun 22, 2008

Comments from Radu Craiu

Dear Nancy,

A few remarks:

1) I fully support SEPARATING Statistics and Mathematics.

2) It is not clear at all how to build units within statistics. For
instance one could build strong arguments about the large overlap
Simulation and Computational Statistics, Applied statistics and
Statistical Methdology (e.g., MCMC could easily fit in all three

3) I see no reason to have Applied Statistics and Statistical
Analysis as
separate units.

4) I see no reason for the Psychometrics as a stand alone
unit. Alternatively, we could have a much larger number of units:
Genomics, Survival models (Biostatistics), etc

5) There should be a Math Stats or Theoretical Statistics unit.

6) One could increase the number of statistical units simply by
more homogeneous units (in terms of problems tackled, methodological
approaches used, theoretical background, etc). Would that be
desirable? See also remark 4.

7) I am puzzled that Math has different units for Classical Analysis
Modern Analysis, Numerical Analysis and Computational Math but
doesn't have separate units for Parametric vs Nonparametric Inference,
Bayesian vs Frequentist. Some of these fields have grown immensely and
may deserve their own units.

Radu Craiu

Nancy Reid said

at 8:19 am on Jun 23, 2008

From Liqun Wang

Hi Nancy,

I agree that we should try to convince NSERC that Statistics is much closer to CS than to Math, and the link CS-Stat is even stronger than CS-OR. We could easily point out many of these links. For example,

> Computer experiment and simulation using parallel/distributed computing is closely linked to CS26;

> Statistical data mining and machine learning is linked to CS29-31 and CS22C;

> Stochastic optimization to MSCS14;

> Data visulization and graphics to CS30-10.

About the building block structure of Statistics, I have the following comments/suggestions:

> I think it makes sence to have two blocks on Theoretical and Applied Probabilities, although sometimes it is difficult to tell the difference.

> I think the block MSCS20: Actuarial Science/Mathematical Finance/Econometrics makes sence. But it is better to change the title "Mathmatical Finance" to "Quantitative Fanance", reflecting the diffrent emphasis of research.

> We could rename the block MSCS18-a as Biostatistics/Bioinformatics/Statistical Genetics.

> The two blocks MSCS18-b Applied Stat and MSCS32 Psychometrics could be merged as one block, which contains also business, inductrial and social statisitcs, etc.

> It makes sence to me to have the block MS16B: Simulation and Comp Stat, although it's covered subjects need to be revised.

> Finally, I think the two blocks MSCS21 Statistical Methodology and MSCS33 Statistical Analysis could be merged and re-split into two or three blocks alonging different lines. For example, Foundation (Theory) and Methodology.

With best wishes,


Liqun Wang, Professor
Department of Statistics, University of Manitoba
Winnipeg, Manitoba, Canada R3T 2N2

Nancy Reid said

at 8:45 pm on Jun 25, 2008

From Sheldon Lin re actuarial science:

Here is my two cents on the changes on actaurial science.

Mathematical finance is now grouped with actuarial sciences under
MSCS20 (was it part of appled math?). In terms of its closeness/relavance
to other areas, I would
rank Operations Research, Applied Math (Numerical Analysis,
Applied Probability, Finance/Economics, Actuarial Science and Statistics
from the most relavant to the least relavant. Also, I think most people in
math finance are employeed in
math and some in industrial engineering. Only a minority are in
departments. This means 1. it might be difficult to review applications
(one needs to find referees outside the traidtional GSC14
refereeing group); 2. there will be applicants who used to apply for
math and applied math. The number of this kind of applicants is likely to
be more than actuarial applicants, which might cause a large increase of
applications. If NSERC really wants to include financial research in
GSC14, applications perhaps should be limited to quantitative finance.

The areas listed under MSCS20 do not seem representing the research areas
well. Game theory and mathematical economics belong to math or
economics. numerical methods perhaps should be replaced by numerical
finance. risk analysis-classical: I really do not know what it means.
actuarial sciences is too vague. Risk
theory is part of actuarial science.
Other areas such as actuarial mathematics, life insurance,
property and casualty insurance, insurance risk management should be
listed. Reliability is
not part of actuarial science nor finance. It belongs to industrial
engineering, OR or management science.


Nancy Reid said

at 8:59 am on Jun 26, 2008

From Mike Evans:

Hi Nancy;

Just a couple of thoughts on this.

1. I think the statistical community should be focused on arguing for an
autonomous committe, even if we know this argument is futile and we would
wind up being placed in with math. To me autonomy is the most scientifically logical outcome,
and the best for STA, and we should argue for that. I realize this might be futile but see 2.

it seems to me that the culture of MAT is much closer
to STA than that of CSC:
CSC is very diverse, e.g., hardware, operating systems, distributed
computing, programming languages, design of algorithms, numerical analysis, computational complexity, machine learning, etc. It isn't
my sense that machine learning (perhaps the sub-area of CSC closest to STA) is really considered
as a high-profile, central sub-area. I think
there is a significant risk that we would be regarded as very marginal
in the context of a joint CSC-STA committee

Nancy Reid said

at 3:10 pm on Jun 26, 2008

From Karen Kopciuk:

Dear Nancy:

I cannot provide a broad perspective on the important issues
you have raised, but can provide a couple of points from my own narrow view
of statistical research.

No single unit seems to capture my research areas; I could likely submit a
grant application to at least three of them for most of my recent
projects. The unit titles do not seem to describe methods
research domains with applications very well.

I also strongly support aligning ourselves with computer scientists,
if we need to be included with another discipline. In my case, there is an
active group of mathematical biologists at the University of
Calgary but it is the bioinformatics researchers with whom I
interact. I have much more in common with them than with any
of the work being carried out by the mathematicians yet we all share a lot
of common application areas.

I do hope we are able to get this restructuring right as the implications
are substantial and likely long term.

Karen Kopciuk

Nancy Reid said

at 1:22 pm on Jun 30, 2008

From Brenda MacGibbon:

Here is my answer to post.

NSERC wants to face the challenges of "the rapid emergence of new
areas and proposals that cross traditional boundaries" by adopting a
conference model. Many statisticians should be happy with this goal,
although we may be unsure of the means proposed to attain it. Statistics
is one of the fields that is emerging as a driving force in fields as
diverse as genetics, biology, environmental science, economics etc.
We are well equipped to play an important role in the emergence of new
areas of research. Computer science is another field that plays a key
role in driving research in many different domains. It seems an ideal
choice to put these disciplines together in the same conference model.
However, NSERC seems to have decided to isolate statistics, by turning the
clock back and merging statistics with mathematics. Many of our finest
statisticians fougt long and hard to free us from this bondage and the
emergence of a stand-alone committee for statistics really improved
statistical research in Canada. I deplore this recent choice of NSERC.

Jinko Graham said

at 4:25 pm on Jul 1, 2008

Some minor suggestions on building blocks in MSSC 18a:

Topic "estimation of selection pressure" could go under either Population Genetics, Evolutionary Genetics or Statistical Genetics

Topics "estimation of breeding values", "QLT analysis" (QTL analysis??) could be grouped under Statistical Genetics or Quantitative Genetics

Topic "phylogenic methods" (phylogenetic methods??) could go under Evolutionary Genetics

Add Genetic Epidemiology

Besides LS17 Quantitative Biology and LS19 Evolutionary Ecology, there are potential links with
* LS16 Systems Biology - for all the Statistical 'Omics
* LS22 Population Biology - for Population Genetics, Genetic Epidemiology and Statistical Genetics
* LS3 Plant Reproduction and LS6 Animal Nutrition and Reproduction - for Statistical Genetics (quantititative genetics and breeding)

Paul Gustafson said

at 5:22 pm on Jul 9, 2008

The push for a stand-alone panel is exactly the right move, I think. Whether it is with granting agencies or university administrators, it seems we must constantly broadcast the message of our inherent interdisciplinarity and our distinctiveness from mathematics. I appreciate the committee's dedication in making the case to NSERC.

With regard to the proposed units, I have a couple of thoughts. I think the distinction and separation between "Biostatistical methods" on the one hand, and "Quantitative methods in Health Science" on the other hand, is a bit dangerous. I say this for a couple of reasons. First, it looks like this would put more theory/methods biostatisticians (tending to be from statistics departments/groups say) in a separate unit from more methods/applications biostatisticians (tending to be from medical and public-health schools). There isn't very good integration between these camps as it stands in Canada, and I would hate for this trend to be exacerbated. Second, the document speaks to links with CIHR and SSHRC, which I agree are pretty crucial in the big-picture. Having a structure where unit #1 clearly maps to CIHR and unit #13 clearly maps to SSHRC could be advantageous in this regard. I imagine the concern with moving the "health" part of #13 to #1 would be unbalanced group sizes.

Another comment is that missing-data, mismeasured-data, etc. does not find a natural home in the current scheme.

Another comment is that I wonder whether the groupings need to be 100% mutually exclusive. As one example, I see `causal models' under #5. I could envision some proposals with causal themes fitting with the general tenor of #5, but then others fitting with the general tenor of #1, for instance.

Again, thanks to the committee for these efforts, and for modernizing the community by making us Wiki-savvy...

Paul Gustafson

Martin Barlow said

at 5:50 pm on Jul 9, 2008

As far as probability is concerned, I will try to consult with the community. My guess is that if Statistics forms its own broad area, many would prefer to be with Math. (I would for example.) If Stats goes in with CS, then almost everyone would want to be with Math.

Martin Barlow

Gail Ivanoff said

at 11:51 am on Jul 10, 2008

I agree with Mike Evans that I think statistics and probability fit far better with mathematics than computer science. In response to Martin Barlow's comment, as a probabiity theorist I will definitely remain with stats provided that stats goes with math. If stats goes in with CS, I will seriously consider moving to Math.

Nancy Reid said

at 2:39 pm on Jul 10, 2008

From Richard Lockhart to NR:

Dear Nancy:

I have spend a wee bit of time trying to comment sensibly on the proposed broad area panel and really failing.

1) I like the proposal you have drafted and agree wholeheartedly with the sentiment that it is counterproductive to force us back in with mathematics.

2) Some of the rationale on offer -- avoiding splitting big disciplines arbitrarily to make suitable sized committees -- simply doesn't apply to statistical sciences. Rather the opposite -- the need for big top level panels is forcing irrelevant and unhelpful mergers.

3) If nearly everyone going to GSC 14 ends up in the MATH grouping won't most of us end up in sub panels for which no mathematician would be a suitable member?

4) But the real problem for me is that I don't understand how it will really work. Each proposal will be reviewed by some set of k people.
They will report to some other set of people. Who will make a decision
and how? It seems to me to be important in view of my comment 3 -- if
the effect for most GSC 14 members is that they get reviewed by a subpanel
of statisticians then not much has really changed except that perhaps a
broad area math panel will consider that everything coming out of stat is
of lower value than real math.

Nancy Reid said

at 2:41 pm on Jul 10, 2008

And NR's reply:


Your view in 4 below is I think quite right, and it is what NSERC keeps saying.
We would simply be a slice of the Math and Stats group, and things would proceed more or less as usual.
There is now a Group Chair for math and stats, so NSERC's view is that it is practically speaking a minor change.
In which case it doesn't make sense to make our pitch.
And, it probably doesn't matter whether we are with math or with cs, we'll be semi-independent in either case. In fact I think we'd be better with math, simply because we'd be a larger fraction of the Group.

However, the budget will be allocated at the Group level, i.e. at the Math and Stats level or CS and Stats or... . It will then be assigned to applicants by the Group Chair and a subcommittee of section chairs. We would have one member on this subcommittee, math would probably have 2, plus there will be a group chair, probably usually from math. These people will not re-evaluate any applications, they will take the ranks from the section evaluations and match $$ to ranks. NSERC will somehow have to be aware of your last point, that the rankings from statistics might get lower value than the rankings from math, either inherently, or because we shoot ourselves in the foot. They are aware of this issue, but haven't really said how they will deal with it.

... continued in the next comment...

Nancy Reid said

at 2:41 pm on Jul 10, 2008

... continued from reply to Richard...

One reason for proposing our own Group is that the optics of the NSERC's version is that we are being folded back into math, which is certainly how many in our community see it, and that is just a bit depressing, it feels like we're moving back, not forward. Personally I think math is moving forward too and it would not be as awful as all that, but I'm not working in a math and stats department, and my research is on the math-y side of stat. Another is that while in the short term we would be a relatively independent, while the institutional memory of the existence of GSC 14 exists, in the longer term with Groups being effectively the new GSCs, we could lose some of this independence. I don't know if this is a sensible thing to worry about or not, it involves prediction under uncertainty! In my view the most important reason for our proposal is that there is a group of 'applied statisticians' for lack of a better descriptor who are not well served under the current model. The largest of these is I think biostatistics, but this feeling of being not theoretical enough for GSC 14 and not applied enough for subject matter GSCs definitely seems to resonate, and I think it would be very good for our discipline to be proactive in trying to solve this. It does mean however that enough people have to buy into this model to make it work, we can't form a separate Group if we don't have a credible number of people who could serve on this Group, and take their place among other Group Chairs. It is this last point that makes me nervous, actually.

Robert Platt said

at 2:53 pm on Jul 10, 2008

I strongly support the proposal that "Statistical Sciences" is an approprate top-level panel. While the discipline is small in terms of number of researchers, the breadth and natural interdisciplinarity that Paul and Richard have both alluded to is the critical issue.

Regardless of the size of the field, we as a community have to push and argue and lobby that statistical sciences should be treated as a distinct and independent unit.

I think Paul's right, too, that this has to be done at the university level too. For example, it's great for other researchers at McGill that there are statisticians "embedded" all over campus (math/stat, epi/biostat, medicine, psychology, management, economics, etc.) but it's not good for statistics itself because none of these groups by itself has the necessary critical mass.

Finally, I think grouping statistics with math or CS will have a negative effect on closer ties between NSERC and CIHR and SSHRC. It's quite possible that this move, regardless of the way groupings within the broad area panel are defined, will cause some researchers at the CIHR-tail-end of the spectrum of statisticians to simply stop applying to NSERC. This would mean more funds for those who continue to apply, but would have negative consequences for statistics.

Nancy Reid said

at 11:33 am on Jul 11, 2008

Suggest blocks 3 (Statistical inference) and 6 (Bayesian inference) be regrouped as:

2. statistical inference
- parametric inference
- Bayesian inference
- nonparametric inference
- inference from stochastic processes
- semi parametric inference
- robust inference
- estimating functions

3. statistical methodology
- Bayesian methods
- Markov chain MC methods
- measurement error
- multi-level and hierarchical datas
- time series analysis
- resampling methods

Nancy Reid said

at 11:35 am on Jul 11, 2008

Suggest "Environmetrics" be called "Quantitative Methods for Environmental Sciences"

and "Spatial Statistics" -> Quantitative methods for spatial data

Nancy Reid said

at 11:37 am on Jul 11, 2008

Suggest 8 (Industrial Statistics) be called "Quantitative methods [or statistical methods] in engineering and industry.

Nancy Reid said

at 2:55 pm on Jul 12, 2008

Suggestions re blocks from Liqun Wang:

Dear Nancy,

I strongly support the proposal of a separate Statistics Group (Broad Area Panel). I have some comments on the units.

First, I agree with Paul that it is better have Biostats and Quantitative Methods in Health Sciences together rather than separated. I therefore suggest to add an item "Quantitative methods in medical and health sciences" in SQI-1.

Secondly, I suggest to modify the title of SQI-13 as "Quantitative Methods in Economics and Social Sciences", and add an item "Econometrics" in this unit. While econometrics and statistics are very close in USA and Europe, it is unfortunate that they are somewhat separate in Canada. I believe that it will be beneficial for both groups if they come closer.

Thirdly, in SQI-9 I am not sure what "Risk Theory" means. It is probably better to change this item to "Risk modelling and management". I also suggest to add some "finance" items such as "First passage time" and "Option pricing" or "Financial derivative evaluation".

Liqun Wang

Nancy Reid said

at 5:04 pm on Jul 15, 2008

From Mary Thompson

Hi Nancy,

I think that the proposal is very much worth pursuing. I was excited to see the draft set of units, which represents the discipline much better than what was originally suggested, and to my mind really takes advantage of the new structure NSERC is adopting, for the benefit of other disciplines as well as ours. I hope NSERC is receptive to something like this.

Re the details, I would add Latent variable models under Psychometrics in SQ13.

In the Complex Survey Analysis section, if it were up to me I would replace "Survey sampling" by "Survey sampling theory and methods", and take out "Survey weights", which would then be subsumed. (Survey weights are not glamorous, and to my mind not a research topic unto themselves.) Perhaps replace "Variance estimation" (also not glamorous) by "Complex survey analytical techniques." Move "Item response theory " to SQI-13 under Latent variable models. It was important to Statistics Canada because of the interests of social sciences researchers using their data, but I think actually belongs to psychometrics.


Nancy Reid said

at 4:05 pm on Jul 17, 2008

From Martin Barlow:

Dear Nancy,
AS its pretty close to your 18 July deadline, and I'm at the IMS
meeting in Singapore, I thought I'd send you my comments on the
Statistics proposals as they relate to Prob. Th.

1. Having one panel covering Prob Th. seems reasonable. Splitting
into 'pure' and 'applied' is difficult and sometimes controversial,
and my guess is that the overall numbers of people are small
enough for one panel.

2. One should consider linking 'Prob Th' with Math Physics, since
Statistical Physics is an important part of Prob. Th, but the
people in that area do want to keep their link with Math. Phys.

3. We don't want Prob. Th. to be split up (as at present) between
different panels.

4. There are mixed views on whether Prob. Th. would be better
placed in Math. or Stat., but it is nearly unanimous that we
don't want to be with CS.

I'm sending this also to Don, who I think is on the Math.
liason committee?


Nancy Reid said

at 4:06 pm on Jul 17, 2008

From Neal Madras

Martin has summed up the comments we have seen pretty well, Nancy.
If Stats goes to CS, then some probabilists will have a tough
choice but most will opt for Math without much hesitation---though
not without regret.


Nancy Reid said

at 10:38 am on Jul 18, 2008

From Helene Massam

Dear Nancy,

Here are a few comments on our GSC restructuring proposal.

I would like to say, first, that I fully support an independent Broad Panel Area
for Statistics. I am glad we moved away from having to choose an affiliation with Math or CS or any other discipline since this would have maintained us as a sub-discipline.

Not having to choose an affiliation with a "foreign" discipline will also allow us to cooperate with many disciplines freely and independently.
Personally, I consider cooperation with mathematics as exciting as cooperation with machine learning.

In order to maintain the impressive recent development of Statistics, we need
to maintain the freedom to choose our research activities without having to consider any particular administrative attachment to another broad area panel.

Best, Helene

You don't have permission to comment on this page.