Thoughts on "Proving" NLP


Intro to NLP
NLP Definition

NLP Books
NLP Tapes

Book of Month
NLP Trainers
NLP Glossary

NLP Reviews
NLP Links

NLP Book Reviews

What's DHE?

International NLP Resources

NLP Trainers

NLP People

NLP Research

NLP & Therapy

NLP & Education

NLP & Health

NLP Training

NLP, IQ, & Genius


NLP was developed in the mid-70s by John Grinder, a Professor at UC Santa Cruz and Richard Bandler, a graduate student. NLP, as most people use the term today, is a set of models of how communication impacts and is impacted by subjective experience. It's more a collection of tools than any overarching theory.

Much of early NLP was based on the work of Virginia Satir, a family therapist; Fritz Perls, founder of Gestalt therapy; Gregory Bateson, anthropologist; and Milton Erickson, hypnotist.


by Stever Robbins

My personal background is in systems theory and large information systems design. I am distressed with DSM-III style categorization for a number of systems-theory reasons (it categorizes the symptoms, not the underlying systemic causes; and categorizations are very subjective "subject appears listless" [Damn! Did that client drop his shopping list again?]).

Part of NLP's attraction to me is that the therapist builds a precise model of the cognitive structure of the client's problem. S/he builds it through minimal nonverbal cues, certain language patterns, and certain eye movements. There is no reliance on interpretations of words like "listless." Two well-trained NLP practitioners should reach identical diagnosis, and be able to tell you exactly how they reached it, down to the eye movement.

Last Thursday, several friends came over to watch a videotape of Richard Bandler doing therapy. We stopped the tape after the everything the client said, and every noticable nonverbal movement. We traced the patterns in her behavior, and analyzed her language in NLP terms to uncover the "significant" imagery that was causing her problem. Richard's intervention dealt with the very things we picked up on. The session included an eight month followup, in which the client said that her problem (panic attacks) had been cured in that one session. Our conclusion: the intervention, which was based on the observable things we picked up on, was successful. Ergo in this case, the diagnostics and intervention worked together to fix the woman's problem.

I believe there are four big problems with NLP, experimentally:

  1. NLP takes skill. You notice minimal nonverbal cues. You notice and respond to linguistic nuances. Intervention-wise, you must do interventions smoothly. If you aren't "congruent," the intervention won't work. It's hard to find experimentors who have trained in the skills and are also objective enough to do credible research.
  2. There's no quality control in NLP. Most of the people out there spouting the lingo are mediocre at best, and often downright incompetent. So depending on which NLP practitioners you deal with, you may well get bad results. [This gets into ethics, too. Quality control and ethics are tightly related!]

  3. It's hard to isolate small chunks of NLP to test. You can't test a diagnostic system independent of its interventions (because a diagnosis is only meaningful insofar as it tells you that certain other actions will have certain effects). And you can't test the interventions unless you're willing to accept the diagnostics, since otherwise you don't know which intervention to use.

    Much NLP research tries to "prove" diagnostics, like the NLP eye movement model. NLP does not say there's an underlying relationship between eye movements and type of cognitive processing. It says if you pretend there is, you will get certain results by using the pretended relationship to guide your intervention. The distinction is critical. In the aforementioned videotape, the client made many eye movements that didn't fit the NLP model. But certain eye movements right before the client said certain things about her problem guided the successful intervention.

    In order to test an intervention, you have to take the diagnostics on faith. This isn't just true of NLP. For example, Newton came up with the wonderful relationship F=MA, force is equal to the product of mass and acceleration. What he spent his time proving was the relationship. He took on faith the existence of force, mass, and acceleration. He didn't try to prove that force existed.

    In NLP, the equation is that eye accessing cues guide intervention. A better thing to test would be: do therapists who use the eye accessing cue model to guide their intervention produce statistically better results than equivalently trained therapists who don't?

  4. People try to use DSM-III diagnostics and NLP techniques. NLP is a way of modeling a client and designing an intervention _for that client_. Much of the testing that I've read about uses one NLP intervention on many people with "the same" problem. But "the same" is defined by a DSM-III style diagnostic, not by the NLP diagnostic. Thus, two people with similar symptoms but different cognitive structures to the problem will react differently to the same NLP intervention.

What we need to test is whether NLP can design a successful intervention for clients more than other therapies. We can ask that question about the population at large, or about specific subsets. (Maybe NLP works better, but only for a certain subset of problems, etc.)

I've been fighting the "is NLP a valid psychological model?" battle for months, and the "real" scientists always turn it into a question of testability, hypotheses, etc. Alas, the four or five times I've asked for help designing a study which would let us know if NLP is even worth studying, only one "real scientist" responded.

Our dialog lasted one round of e-mail:

Scientist: We find out if NLP is valid by comparing a group of people treated with NLP technology with people treated with accepted, "valid" treatments.

Stever: Central to NLP is the skill level of the therapist. A lousy NLP therapist will not get results. The model and skill set require a well-defined, but difficult level of competence. [Like any complex skill.]

How do we make sure that the group of people treated with NLP are treated by competent NLPers? After all, since NLP isn't an accepted field, it's been nearly impossible to get a good credentialing system in place, so just going on people's word "I'm good" isn't going to give high quality results.

The same problem also applies to the "traditional" therapists in the control group.

There was no response from "scientist."

There's a subtler issue at work, too. NLP principles are compatible with all kinds of other tehcniques. A therapist may use another therapeutic model to guide their approach, yet unconsciously be establishing rapport and reacting to client nonverbal behavior in a manner consistent with the NLP models. (After all, the NLP models were developed by observing excellent therapists in other domains.) It's unclear how you build in controls for that situation.

There are a few fundamental problems with using the scientific method on psychology. First, people are HUGE black boxes. It's hard to isolate very much about that. And to correlate observable externals with internal states relies heavily on self-description. Self-description is one of the least accurate gauges of mental state I think I've ever seen. When was the last time someone who felt threatened by a member of the opposite sex actually came out and said it, without an hour's careful rapport & discussion?

That's one problem with testing NLP, by the way. The various NLP diagnostics and tools all work TOGETHER. It's a more holistic model in many ways than most psych [assuming you train with the right NLP trainer, that is]. You can't use an NLP intervention with a traditional psych diagnostic (or vice versa). In order to test the any holistic system, you have to accept the system's distinctions on faith. They're part of the DEFINITION of the system; they aren't part of what you test.

We do this all the time in the traditional sciences, but people like to pretend we don't. For example, we immediately accept the concept of "gravity," because we directly experience it and have partitioned it out in our experience to the point where we've given it a word. We don't spend our time proving that gravity exists; we take it as a given and then find out if it's a useful distinction that we can draw predictive rules from. We discover it is, and physics contains that set of rules.

In NLP, the phenomena used to understand a psychological event weren't commonly observed enough to get their own words. So everyone's all caught up in arguing over whether this or that non-verbal behavior "really" exists. We can never answer that; some kinds of existence are more a function of language than the external world. (Do the seven colors of the rainbow "really" exist? Not in languages that have only three words for color.) The more interesting question, which we're never getting to, is: if we adopt that way of defining the world, can we then go out and produce repeatable results with it?

A lot of psychological science is also done statistically. Well, statistics are great if you're going to give 100,000 people Prozac and figure out than N% will be happy, M% will still be depressed, and Z% will kill themselves over their inability to open the child proof cap. But that doesn't give us any knowledge whatsoever about how to understand a unique individual standing in front of us. Will they fall into the N%, M%, or Z% group? Or maybe some fringe group that was eliminated as statistically insignificant in our 100,000 person study. We don't know. But to them, and to the issue of helping them, it's a crucial question!

Copyright © 1996, Stever Robbins. If you'd like to republish or quote part of this page, just write and we can work something out. Stever Robbins

Back to NLP home page.

© 1993-2008, by Stever Robbins