read:

Measurement issues

Kidder and Judd (1986) differentiate four types of measurement scale - nominal, ordinal,

interval and ratio - each with implications for the most appropriate methods for data

analysis. For example, techniques like regression analysis depend on numerically ordered

data and are unsuitable for ordinal variables, which convey categories rather than levels

(e.g. manufacturing, retailing and financial services). Analysis of variance (ANOVA) is

preferred in such circumstances. Confusingly, Likert scales are strictly ordinal variables,

but they are usually considered as interval variables for analytical purposes because of

their relatively large number of categories (i.e. usually five or seven in practice) and the

incidence of averaging.

Nominal - mutually exclusive and collectively exhaustive categories

conveying no ordering message (e.g. male/female;

manufacturing/retailing/other). Where these are treated as other than

(0, 1) variables, care must be exercised in their use as independent

variables in a regression analysis, since ordinary least squares (OLS)

analysis will treat them as ordered variables. The most common test of

statistical significance for use with this data is the chi-squared ( ) test.

Phi and Cramer's V are the most appropriate measures of association.

2

Ordinal - mutually exclusive categories which can be ordered (e.g.

large/medium/small). Rank-order methods (e.g. Spearman's 'rho') are

the most appropriate measures of association for this data, and

measures of statistical significance are confined to non-parametric

methods (e.g. the Mann-Whitney U-test). Although the use of

parametric methods (e.g. t-test, F-test) is theoretically incorrect for

ordinal data, some researchers will still adopt these techniques on the

grounds that the difference in outcomes is miniscule. Together, nominal

and ordinal variables are often called non-metric variables.

Interval - mutually exclusive ordered categories where specific

intervals have the same meaning but no ratio relationship exists (i.e. a

score of 2.0 is not double the size of a score of 1.0). Thus, a temperature

of 30 degrees is not twice as hot as 15 degrees; a Z-score of 4.0 is not

four times better than a Z-score of 1.0. The product-moment coefficient

of correlation (Pearson's 'r') can be used here to measure association,

and parametric tests of significance are employed.

Ratio - continuous data, where specific intervals have the same

meaning, and multiples have the same meaning (e.g. age, height, weight,

dollars). Again Pearson's 'r' is an appropriate measure of association and

parametric significance tests can be used. Interval and ratio measures,

together, are often called metric variables.

Within our ordinal level measures, alternative rating scales are frequently employed in

the accounting literature:

Linear (sometimes called graphic scales) - e.g. strongly

agree/disagree and points in between these extremes, as long as they

are not labelled verbally. But respondents may be unwilling to select the

extremes on a continuous scale (i.e. a 1 to 7 scale may induce subjects to

generate responses confined to the 2 to 6 range).

Itemised (sometimes called categorical scales) - as above, but with

labels to denote specific ordered categories, i.e. Strongly Agree, Agree,

Neutral, Disagree, Strongly Disagree. The word 'Neutral' is often

replaced in practice by the phrase 'Neither Agree nor Disagree'.

Comparative - measures ask for judgements to be made with reference

to specific levels of performance with the objective of providing a

comparative base.

Multiple-item (of which the Likert scale is much the most commonly

adopted) - the numerical scores on the Likert scale permit items

measuring the same construct to be added. Doing so facilitates the

investigation of the impact of individual items and sub-groups as well as

any incidence of multicollinearity.

Semantic differential - following Osgood et al. (1957), who developed a

semantic differential to measure individuals' perceptions of the meaning

of different terms. A set of seven-point bipolar scales allows

respondents to rate concepts between the extremes of good/bad,

passive/active, positive/negative, etc. The method has been employed in

the accounting environment (most notably by Houghton, 1987, 1988) to

measure understanding of accounting terminology.

Questions to answer: please donot copy and paste the answers

3. Kidder and Judd (1986) differentiated four types of measurement scale. Each with implications for the most appropriate methods for data analysis. Please list and explain all four in details.

The four types of measurement scale are nominal, ordinal, interval, and ratio.

Nominal: is a method that categorizes the research subjects into similar groups. Example, male/female, manufacturing/retailing/other. These observations can be named without a particular order.

Ordinal:

Interval:

Ratio:

4. Within the ordinal level measures, alternative rating scales are frequently employed in the accounting literature. List and explain the five that was mentioned in the book.