In the field of Physical Activity and Exercise Behavior, there have been numerous attempts to apply the TPB. Because the TPB was originally the TRA, designed to explain once off behaviors (voting) it did not do a good job of explaining long term behavior or repeat behaviors. With the expanded TPB incorporating perceived behavioral control (PBC), the model has done a better job of explaining and or predicting behavior. Within the context of exercise and physical activity behavior there have typically been associations between intentions and physical activity as well as between the attitudes and Intentions, PBC and intentions and PBC and Behavior. Subjective norms seems not to be associated as strongly with intentions. An exception was a study with young 9th and 10 grade students who may have relied on the perceived importance of their behavior in the eyes of significant others (teachers and parents) for guidance so developmental stage may be important. Overall there are better theories for examining PA and Exercise behavior that take into account reciprocal relationships such as Bandura's Social Cognitive Theory or even Deci and Ryan's Self Determination Theory.

I second Charles' thoughts on TPB. Modern thinking about behaviour change is begeining to acknowledge the learnings from the behavioural sciences that show that many of our behaviours are implicit and require changes in context and psychological flexibility. Having looked at many TPB studies there is a consistent positive variance (circa 20-30%) around intentions predicting behaviour. This tends to increase when you add motivational measures and more cognitive inititives. However, the bottomline is that intentions whilst being important in predicting change are becoming outdated, overused and insufficient to understand the factors behing actual change in behaviour. It was always felt that the model was flexible enough to add other constructs but I am beginning to wonder how relevant it now is it terms of modern thinking and the economics of behaviour change

There's little doubt that the TPB (which I now refer to generally as the Reasoned Action Model following Fishbein & Ajzen's 2010 book) falls short of explaining variance. And the intention-behavior relationship in the RAM has, of course, always been fuzzy. As a general theory that tries to explain all human behavior, it suffers in part from the sheer scope of what it tries to accomplish, as do other general theories of behavior. However, from an applied perspective when your concern is not to explain and predict, but rather to apply the rational inherent in the theory (which mainly means its expectancy-value logic) in order to identify messages for persuasive communication, I find the RAM a useful and successful framework in terms of measurable behavioral impact. As Charles mentioned, it is parsimonious and easy to operationalize. We've found that if one applies the entire methodology, beginning with a rigorous elicitation to identify behavioral, normative and control beliefs salient to a particular well-defined behavior, and then follows with quantitative measures of the modal salient beliefs among independent samples of known compliers and known non-compliers, even simple mean comparisons between the samples can identify one or more beliefs with the greatest persuasion potential. It is obvious to most (as our colleagues above have rightly pointed out) that these "rationale" beliefs will not capture the universe of factors that influence behavioral decision making. But from an applied perspective, the question is whether these identified beliefs play a predominant enough role to guide a behavioral decision--despite the fact that they fall short of capturing every influence. Our research suggests they often do. In a study of travelers' philanthropy in Galapagos, messages targeting key beliefs resulted in a quadrupling of donatiions in a single tourist season. In another study, messages increased litter pick-up by national park trail walkers by 21% over a control condition. In another study, bird-feeding by first-time visitors to a national park was effectively eliminated. And in a fourth study, more than half of visitors to a national park who intended to drive their private vehicles into the park were persuaded by one of our targeted messages to take the park shuttle bus instead. Of course, the effectiveness of a message will depend on how it's crafted and delivered (too many studies seem to assume that if you simply "say" the belief that somehow a persuasive impact will result), but we have found that if the salient beliefs found to have persuasion potential are communicated in a compelling way (those with high elaboration likelihood, of instance), a significant increase in a target behavior will often result. What we know from these studies is that a RAM perspective helps the design of persuasive communication interventions (at least those aimed at one-off behaviors) and they can lead to appreciable increases in a desired behavior. This is tantamount to saying that targeting salient beliefs about a behavior is a useful approach from an applied perspective. By the way, the RAM has indeed been extended recently. In 2010, Fishbein and Ajzen added Cialdini's descriptive norm to subjective norm to reformulate normative influences as "perceived norm." But measurement of the descriptive norm variable remains a subject in need of discussion. My $0.02.

Well, I have used Theory of Planned Behavior (or Reasoned Action Model as Sam preferred) is a variety of behaviors and I have seen it used by other researchers. In most cases, some aspect of the model can not be proven significant in influencing the intention (much less behavior). My opinion is that the weakest part of the model is the perceived behavioral control, perhaps due to less uniform understanding of the construct by the researchers (much less the respondents).

I also agree with Alexander that the model does poorly in predicting long term behavior. Perhaps some group of planned behavior can be predicted better than other groups of behavior using this model. Thus perhaps in alignment with Sam's view that it suffers from the sheer scope of what it tries to accomplish, there should be better definition on what kind of behavior does the model can explain the best.

I could numerate some limitation for applying TPB as a framework to explain human behaviour as follow:

TPB is a unidirectional model

The model relies on cognitions and omits other potentially important determinants of action, such as environmental influences

The model predicts behaviour from measures of behavioural intention taken at one point in time

Insufficient attention has been paid to the measurement of behaviour

The TPB investigates the interrelationships between model constructs and a single behaviour. It does not account for alternative behaviours