Discussion: View Thread

We are missing an important point to being a successful consulting statistician.  You need to develop an understanding (knowledge) of the process the researcher is working with in order to provide appropriate statistical help.  If a physician is working with a specific cancer drug you need to understand the biology, chemistry and their interaction when treating a subject before you can provide appropriate discussion on the design and analysis of a study.  I have seen many "consultants" OK the design the researcher suggests and the analysis the researcher has seen used in the past.  The researcher's thinking is if someone used this design and analysis in the past and their paper was published then that is what I want to do.  Ofter one misses factors for blocking, etc, when they do not understand the process and there may be better designs and better ways to carry out the analyses.  They will still get their paper published if they carry out a good study.    So how many years of experience do you need?  If you have a degree in Statistics or Biostatistics with a concentration in applications AND you have a background in the science area you are wanting to consult in, then you could do solo consulting (with the backup of a mentor) with little additional experience.  The trick is knowing researchers who are working in the science area you have experience with and convincing them you can contribute to the projects.  Without having a name for yourself through publications or through consulting activities it is difficult to get clients.  If you know of someone or a group of researchers that work in your non-statistical areas of expertise then you will have a head start.  If you do not have experience in another area other than statistics then it will take a few years of consulting with a mentor to build the expertise in areas you wish to provide statistical expertise.
-------------------------------------------
George Milliken
-------------------------------------------

Great advice George.  I agree 100% with the advice.  I don't agree with the comment that we were missing the point.  Understanding the problem thoroughly through a careful dialog with the client requires some familiarity with the subject area.  We can't be experts in every scientific discipline that we want to consult in but we do at a minimum need to brush up on the existing literature in that subject area and get the client to explain to us what we don't know.
-------------------------------------------
Michael Chernick
Director of Biostatistical Services
Lankenau Institute for Medical Research
-------------------------------------------

George Milliken describes first-rate collaborative stat science.
Michael Chernick describes first-rate stat consulting.

The distinction is clear to me.

When consulting, I am fully engaged during the meetings and when I am programming or writing something, but the project does not penetrate my time beyond that.

When collaborating, I "walk around" with the problems and urge myself to be cutting-edge and creative in all phases of my work. I come to understand the subject-matter well enough that I could be at a dinner party with lay people and make them think I was a bona fide expert.

In the real world of academic biostatistics today, most of the problems become consulting ones, because there is insufficient funding to support real collaboration. It is now common to negotiate a funding level of X% to support a collaborative effort, but the actual funding ends up being X/3%. With a promise of X% funding, you proposed to provide Grade A custom analyses and Tufte-esque graphics. But at X/3%, your effort is forced down to becoming a solid Grade B, at best.

Yet this ain't so bad: a solid consulting effort by a Grade A stat scientist is still way better than what is typically done and reported by those with little or no expertise and experience in our discipline.

-------------------------------------------
Ralph O'Brien
Case Western Reserve University
-------------------------------------------







Original Message:
Sent: 09-20-2011 11:25
From: Michael Chernick
Subject: Solo Consultant


Great advice George.  I agree 100% with the advice.  I don't agree with the comment that we were missing the point.  Understanding the problem thoroughly through a careful dialog with the client requires some familiarity with the subject area.  We can't be experts in every scientific discipline that we want to consult in but we do at a minimum need to brush up on the existing literature in that subject area and get the client to explain to us what we don't know.
-------------------------------------------
Michael Chernick
Director of Biostatistical Services
Lankenau Institute for Medical Research
-------------------------------------------

I agree with Ralph.  Sometimes we just don't have the time to do "first class" work and yet I would have no problem doing what Ralph calls Grade B work.  What concerns me is when a person commit a type III error (by my definition "providing a great analysis to the wrong problem").  Under any circumstance we need to understand the problem well enough to know that we are giving a correct solution.  Mindless number crunching is not consulting and should be avoided.

-------------------------------------------
Michael Chernick
Director of Biostatistical Services
Lankenau Institute for Medical Research
-------------------------------------------







Original Message:
Sent: 09-20-2011 12:07
From: Ralph O'Brien
Subject: Solo Consultant

George Milliken describes first-rate collaborative stat science.
Michael Chernick describes first-rate stat consulting.

The distinction is clear to me.

When consulting, I am fully engaged during the meetings and when I am programming or writing something, but the project does not penetrate my time beyond that.

When collaborating, I "walk around" with the problems and urge myself to be cutting-edge and creative in all phases of my work. I come to understand the subject-matter well enough that I could be at a dinner party with lay people and make them think I was a bona fide expert.

In the real world of academic biostatistics today, most of the problems become consulting ones, because there is insufficient funding to support real collaboration. It is now common to negotiate a funding level of X% to support a collaborative effort, but the actual funding ends up being X/3%. With a promise of X% funding, you proposed to provide Grade A custom analyses and Tufte-esque graphics. But at X/3%, your effort is forced down to becoming a solid Grade B, at best.

Yet this ain't so bad: a solid consulting effort by a Grade A stat scientist is still way better than what is typically done and reported by those with little or no expertise and experience in our discipline.

-------------------------------------------
Ralph O'Brien
Case Western Reserve University
-------------------------------------------

Let me give a rough statistic (!) to make my point. About 15 years ago, when I headed the consulting/collaborative biostat group at a large "academoid" medical center, I looked at the number of hours spent per project (N > 700/year). The median was about 10 hours, a number that indicates "consulting." I don't recall where the 80th percentile was, but I'd guess it was under 20 hours. This outstanding group ran on mostly MS-level talent and only the very weakest among them did what their team leaders and I considered to be pedestrian work, i.e., "number crunching." They were actually valuable team members, because they took on the most pedestrian research projects, turned them around quickly doing them well-enough (C+, B-) to "pass" in the medical literature, and kept those kinds of investigators (and their department chairs) happy. This allowed time for the better statistical scientists on the team to work on those projects that truly needed a more collaborative effort. 

To use a football metaphor, the offensive guards and tackles may do the yeoman grunt work, but ask any quarterback or running back how valuable they are.

-------------------------------------------
Ralph O'Brien
Case Western Reserve University
-------------------------------------------







Original Message:
Sent: 09-20-2011 12:18
From: Michael Chernick
Subject: Solo Consultant

I agree with Ralph.  Sometimes we just don't have the time to do "first class" work and yet I would have no problem doing what Ralph calls Grade B work.  What concerns me is when a person commit a type III error (by my definition "providing a great analysis to the wrong problem").  Under any circumstance we need to understand the problem well enough to know that we are giving a correct solution.  Mindless number crunching is not consulting and should be avoided.

-------------------------------------------
Michael Chernick
Director of Biostatistical Services
Lankenau Institute for Medical Research
-------------------------------------------







Original Message:
Sent: 09-20-2011 12:07
From: Ralph O'Brien
Subject: Solo Consultant

George Milliken describes first-rate collaborative stat science.
Michael Chernick describes first-rate stat consulting.

The distinction is clear to me.

When consulting, I am fully engaged during the meetings and when I am programming or writing something, but the project does not penetrate my time beyond that.

When collaborating, I "walk around" with the problems and urge myself to be cutting-edge and creative in all phases of my work. I come to understand the subject-matter well enough that I could be at a dinner party with lay people and make them think I was a bona fide expert.

In the real world of academic biostatistics today, most of the problems become consulting ones, because there is insufficient funding to support real collaboration. It is now common to negotiate a funding level of X% to support a collaborative effort, but the actual funding ends up being X/3%. With a promise of X% funding, you proposed to provide Grade A custom analyses and Tufte-esque graphics. But at X/3%, your effort is forced down to becoming a solid Grade B, at best.

Yet this ain't so bad: a solid consulting effort by a Grade A stat scientist is still way better than what is typically done and reported by those with little or no expertise and experience in our discipline.

-------------------------------------------
Ralph O'Brien
Case Western Reserve University
-------------------------------------------

The Cleveland Clinic is an exceptional group.  I wouldn't consider their work typical.  The graduate student that does volunteer work for me does an exceptional job with only a bachelor's degree and some mentoring from me.  One person on this thread said that just running routine analyses for physicians would be consulting work that an inexperienced but well-educated statistician can do.  I contend that such work is not consulting.  Consulting requires rendering expert advice.  My other contention is that physicians often don't know exactly what they need and so it takes an exchange between the consultant and the physician (could be merely one or two half hour sessions).  But if a physician directs you to compute means, standard deviations and confidence intervals for X, Y and Z and you just go right along and do it with no questions asked then you are doing mindless number crunching (not consulting or collaborating).

-------------------------------------------
Michael Chernick
Director of Biostatistical Services
Lankenau Institute for Medical Research
-------------------------------------------







Original Message:
Sent: 09-20-2011 13:12
From: Ralph O'Brien
Subject: Solo Consultant


Let me give a rough statistic (!) to make my point. About 15 years ago, when I headed the consulting/collaborative biostat group at a large "academoid" medical center, I looked at the number of hours spent per project (N > 700/year). The median was about 10 hours, a number that indicates "consulting." I don't recall where the 80th percentile was, but I'd guess it was under 20 hours. This outstanding group ran on mostly MS-level talent and only the very weakest among them did what their team leaders and I considered to be pedestrian work, i.e., "number crunching." They were actually valuable team members, because they took on the most pedestrian research projects, turned them around quickly doing them well-enough (C+, B-) to "pass" in the medical literature, and kept those kinds of investigators (and their department chairs) happy. This allowed time for the better statistical scientists on the team to work on those projects that truly needed a more collaborative effort. 

To use a football metaphor, the offensive guards and tackles may do the yeoman grunt work, but ask any quarterback or running back how valuable they are.

-------------------------------------------
Ralph O'Brien
Case Western Reserve University
-------------------------------------------







Original Message:
Sent: 09-20-2011 12:18
From: Michael Chernick
Subject: Solo Consultant

I agree with Ralph.  Sometimes we just don't have the time to do "first class" work and yet I would have no problem doing what Ralph calls Grade B work.  What concerns me is when a person commit a type III error (by my definition "providing a great analysis to the wrong problem").  Under any circumstance we need to understand the problem well enough to know that we are giving a correct solution.  Mindless number crunching is not consulting and should be avoided.

-------------------------------------------
Michael Chernick
Director of Biostatistical Services
Lankenau Institute for Medical Research
-------------------------------------------

Let me add on to the comment about working with physicians after 35+ years.

Most (of course, not all) researching physicians are very weak when it comes to mathematics. One school of thought is that physicians are "concrete thinkers", who like to be able to see and touch something for it to be real.  Math types, on the other hand, tend to be "abstract thinkers" for whom scribbles on a blackboard are just as real as the chair they are sitting on, but not for concrete thinkers.  For this reason, physicians can say all the right words about statistics in research, but have little comprehension of it, if challenged.

I tell people that physicians take a lot of science in school, but they are not scientists. A scientist has to grasp the "big picture", which many have a hard time doing.  So, don't be surprised if some rudiments of design and analysis seem to be hard for them to grasp.  They do know that p<0.05 is good, but will struggle to explain what it indicates.  So, don't be surprised when they ask for a power analysis when doing a simple reliability study.  They are wonderful care givers, but we live in a different world than they do.

-------------------------------------------
Richard Browne
Texas Scottish Rite Hospital for Children
-------------------------------------------







Original Message:
Sent: 09-20-2011 13:35
From: Michael Chernick
Subject: Solo Consultant

The Cleveland Clinic is an exceptional group.  I wouldn't consider their work typical.  The graduate student that does volunteer work for me does an exceptional job with only a bachelor's degree and some mentoring from me.  One person on this thread said that just running routine analyses for physicians would be consulting work that an inexperienced but well-educated statistician can do.  I contend that such work is not consulting.  Consulting requires rendering expert advice.  My other contention is that physicians often don't know exactly what they need and so it takes an exchange between the consultant and the physician (could be merely one or two half hour sessions).  But if a physician directs you to compute means, standard deviations and confidence intervals for X, Y and Z and you just go right along and do it with no questions asked then you are doing mindless number crunching (not consulting or collaborating).

-------------------------------------------
Michael Chernick
Director of Biostatistical Services
Lankenau Institute for Medical Research
-------------------------------------------







Original Message:
Sent: 09-20-2011 13:12
From: Ralph O'Brien
Subject: Solo Consultant


Let me give a rough statistic (!) to make my point. About 15 years ago, when I headed the consulting/collaborative biostat group at a large "academoid" medical center, I looked at the number of hours spent per project (N > 700/year). The median was about 10 hours, a number that indicates "consulting." I don't recall where the 80th percentile was, but I'd guess it was under 20 hours. This outstanding group ran on mostly MS-level talent and only the very weakest among them did what their team leaders and I considered to be pedestrian work, i.e., "number crunching." They were actually valuable team members, because they took on the most pedestrian research projects, turned them around quickly doing them well-enough (C+, B-) to "pass" in the medical literature, and kept those kinds of investigators (and their department chairs) happy. This allowed time for the better statistical scientists on the team to work on those projects that truly needed a more collaborative effort. 

To use a football metaphor, the offensive guards and tackles may do the yeoman grunt work, but ask any quarterback or running back how valuable they are.

-------------------------------------------
Ralph O'Brien
Case Western Reserve University
-------------------------------------------







Original Message:
Sent: 09-20-2011 12:18
From: Michael Chernick
Subject: Solo Consultant

I agree with Ralph.  Sometimes we just don't have the time to do "first class" work and yet I would have no problem doing what Ralph calls Grade B work.  What concerns me is when a person commit a type III error (by my definition "providing a great analysis to the wrong problem").  Under any circumstance we need to understand the problem well enough to know that we are giving a correct solution.  Mindless number crunching is not consulting and should be avoided.

-------------------------------------------
Michael Chernick
Director of Biostatistical Services
Lankenau Institute for Medical Research
-------------------------------------------