Talk:Validity (statistics)

From Wikipedia, the free encyclopedia

[edit] Moved note re:Criterion validity

Anon 142.103.116.65 left the following note in the article space for Criterion validity, red-linked from this article. Since the note is more appropriate for a talk page, and that article doesn't exist yet, I'm moving the note here before deleting the article.

Criterion validity is not the right terminology. It ought to be "criterion related validity" which means the validity is actually of the predictor using the specific criterion. It is safer to term it "predictive validity" of a measure.

End copied text. SWAdair | Talk 07:14, 11 Mar 2005 (UTC)

As far as I know "A test can be reliable, but not valid" is the right definition or description. Imagine a clock shows every day 3.00pm, it is everyday the same, therefore it is reliable. However, if you try to measure your weight with a clock, you have a reliable measurement without validity.

[edit] Validity and reliability

The article states that A valid measure must be reliable, but a reliable measure need not be valid. , but Earl Babbie's 'The Practice of Social Research', 10th edition, p.145 has a graph that implies that a valid measure does not have to be reliable. Can anybody elaborate on this? --Piotr Konieczny aka Prokonsul Piotrus ^Talk 18:47, 16 October 2005 (UTC)

What is meant is that if measurements of a person's weight are to be valid (i.e. they actually measure weight) they must me reliable. They cannot change from instrument to instrument etc. However, an instrument can give consistent measurements - hence be reliabile - yet not measure what it is supposed to measure. Thus, it would be 'reliable' but not valid. This is a common argument. The problem with it is that the distinction between validity and reliability is blurry at best. It is taken for granted that validity means something measures the trait or attribute it purports to measure. Generally, people implicitly take it that something is only reliable if it measures what it purports to, and if so, the statement you cited ceases to make sense. It depends on how reliability is defined, in precise terms. Holon 02:00, 1 April 2006 (UTC)

I'm not sure I agree with you. Most textbooks and articles in both psychology (e.g. D. Borsboom et al., "The concept of validity". Psychological Review, 111,4,pp. 1061-71) and the social sciences (e.g. King, Keohane and Verba's well-known textbook) define validity and reliability clearly as separate concepts. Neither one necessarily implies the other. The classical explanation of this view is that of a rifle pointed at a target; a rifle aimed exactly at the bull's eye represents a valid measurement. But it may still be off because of random errors (imprecision). On the other hand, another rifle may be very precise (reliable) but pointed somewhere else completely, and thus invalid if you want to hit the bull's eye. A more statistical formulation is that unreliability is about random error, while invalidity is about systematic error. Your statement that "validity means something measures the trait or attribute it purports to measure" is indeed common, and can be expanded with "but not necessarily with perfect precision". 84.76.46.242 16:02, 4 November 2006 (UTC)

Reliability and validity are related but independent. They are analogous to the engineering terms precision and accuracy respectively. An analog wristwatch that does not work is accurate (valid) twice a day to as many decimal places as you can measure. But it lacks precision (reliability). A watch than is always 10 minutes fast is never accurate but is very precise. These terms are well defined and accepted in engineering.

The problem comes in when mapping these concepts into social science because the terms acquire linguistic uncertainty from colloquial usage. In every day usage for example, a reliable person is always on time. Using the scientific definition of reliability, a person that is always 10 minutes late is also reliable.

[edit] Needs Rewrite for Clarity

• I would like to see this expanded using this outline as a guide:

I. Validity
 A. Internal
 B. External
 C. Statistical Conclusion
 D. Construct
   i. Intentional
   ii. Representation
    a. Face
    b. Content
   iii. Observation
    a. Predictive
    b. Criterion
    c. Concurrent
    d. Convergent
    e. Discriminant