Overview

Steps in measurement

An illustration

Level of measurement

Validity

Reliability

Relationship between
validity and
reliability

Reactivity

Ethical considerations

Aids

Once the inter-item correlation coefficients are found, the remaining steps are easy. The mean for all the correlation coefficients is found by adding the values for all the coefficients and dividing by the number of coefficients.   The result, called alpha (?, a Greek letter) is put into the following formula for the calculation of alpha, the Chronbach measure of reliability.

Assume that our calculation of the mean coefficient of correlation was .55, which would represent a moderate degree of relationship, on the average, between each item and every other item in the scale. Putting the value of .55 for ? and the value of 10 for the number of items into Chronbach's formula, we get:

                n(?)            10(.55)            5.50               5.50              5.50
alpha = ________ = ___________ = ________ =      _______    =   ______   = 0.92
             1+?(n-1)     1+ .55(10-1)     1+ .55(9)         1 + 4.95           5.95

Like a positive correlation coefficient, alpha varies between 0 and 1.00. Higher values indicate greater reliability. In this example, an alpha of 0.92 would indicate a high estimate of reliability.

For an additional discussion of reliability, visit Types of Reliability. This discussion covers all ways of establishing reliability and provides a technical and precise definition of reliability based on statistical operations.

Relationship between validity and reliability

We have presented validity and reliability as separate concepts so that you understand each by itself. In research, we want measuring instruments that are high on both validity and reliability. But in practice, we may find any of four relationships between validity and reliability:

• High reliability, but low validity — the indicators measure something consistently but not the intended concept;
• High validity, but low reliability — the indicator represents the concept well but does not produce consistent measurements;
• Low validity and low reliability — the worst case; the indicators neither measure the concept nor produce consistent results of whatever they measure; or
• High validity and high reliability — what we hope for; the indicators consistently measure what they were intended to measure.

When poor validity or reliability results occur, a researcher has to seek the source of error, revise the measuring instrument as appropriate, and test again for validity and reliability.

Reactivity

Reactivity arises when the method used to obtain data affects the measurement of some variable. This problem is particularly acute in the social sciences because humans can and do react to being observed or being asked questions. Knowing they are being observed, some people will react by becoming self conscious and even change their behavior to look good in the eyes of the observer. In a survey, for example, respondents may give socially approved answers instead of truthful ones for certain questions.   In both instances, invalid and probably unreliable data would be obtained which would put any findings from the investigation in doubt. Any conclusions from the study would probably be invalid as well.