I think it really depends on the context in which you are using these measures. If you ask them in the context of a survey interview, you will probably get similar responses each time, which would make them reliable in a technical sense. But if the question is whether they would predict actual behavior in some real world setting, that is another matter -- which should probably fall under the heading of validity rather than reliability.

Behavior can be very context dependent, so the question of whether you can predict behavior from attitudinal measures has to consider the circumstances in which you are trying to make that prediction. In other words, attitudes of all sorts are typically measured in a very general fashion, while behavior always occurs in specific conditions.

Dear Dr. Morgan,

Thank you very much for the astute response - very helpful!

Yours,

David

Dr Morgan makes two very good points.  

In regard to validity,  some types of questions are reasonably good predictors of likely behaviour  over a  population (voting polls for example), though not for individuals (M. Desroaches makes a very important point). Unfortunately, understanding  the behaviour is much more problematic. For other issues, direct questions of any kind, are a very POOR predictor of behaviour (see La Pierre's study of racism in the 1940s) and even worse at recall (see Brewer 2012, or Loftus 1974,1975, etc. on eyewitness testimony, false memory and recovered memory).

In relation to reliability, as long as the questionnaire is the same and administered in the same way, to a similar population you are likely to get similar results.

However, changes to a questionnaire may result in different answers.  Context dependency applies to survey questions as well to behaviour. There is very good evidence that slight changes in the order of questions or the addition of 'priming' questions (see Tourangeau), or even changing the values in a scale response item (see Schwarz and Hippler)  can modify the results. 

Survey researchers tend to see such effects   as 'errors': but 'recent' developments in cognitive science suggest there may be more fundamental issues with our understanding of what is happening when we ask questions in a research interview or questionnaire. Tourangeau (1987) proposed a model based on a logical process for responding to a question.  The model now appears inadequate mainly because most of the cognitive processes involved are not based on 'logic' but are quite literally responses'  First, as Kahneman (2011) points out  participants are almost certainly answering a different question, indeed a set of multiple questions, to the one asked of them.   Second, it seems likely that judgements, attitudes, even memory, are reconstructed at the time of use and are thus context dependent.  I suggest we probably do NOT 'hold' attitudes or 'store' memory but create them from patterns in the brain and 'express' them in context.  Third,  'judgements' come before memory and our recall tends to confirm our judgement ('confirmatory information processing') unless we make great effort to test our judgements. Fourth,  we express different attitudes in different contexts. It seems likely that we 'hold' multiple attitudes and views rather than a stable, fixed set of attitudes, etc. Fifth, research over many years into 'projective' or 'elicitation' techniques in market research, shows that they can be better predictors of buyer behaviour than questionnaires (Haire 1950; Breivik et al 2003).  Sixth, there seems to be a strong element of self-presentation in responses to direct questions.  This is glossed by survey researchers as 'social desirability bias' but as schwarz and Hippler's research and the other five points above suggest, it is probably more complicated than a simple bias.  I could go on but I think I have made the point.

Nevertheless, individuals do appear to be have recurring patterns of responses, in similar contexts. That is,  an individual makes certain responses more frequently than other responses.  So we might postulate that some responses are more probable than others, depending on context.  (There is some synergy here with Jungian notions of preferences and your use of the term predispositions.)

The cognitive information processing model and the schema concept between them suggest a model for answering questions that might prove fruitful.  See my papers on this site for more on that idea. I hope to produce more work shortly.

I believe it depends on the validity and reliability of the data you collected which arise from the validity and reliability of your data-collection instrument.