Staffing Organizations (6^th ed.)

“Measurement”

Chapter 7 Notes

The existence of inter-individual differences means that, for any particular KSAO (knowledge, skills, abilities and other attributes), some individuals will be more qualified (better matched to job requirements) than will others. Intra-individual differences show that, since not all jobs have the same KSAO requirements, any given individual will be more suited to perform some jobs than others. Measures of KSAO characteristics of individuals are referred to as predictors, or tests. The related work behaviors or characteristics are called the criterion. Criterion measures quantify outcomes.

· Applicant flow statistics require the calculation of selection rates for the groups under analysis.

· Applicant stock statistics for groups under analysis require calculation of percentages for availability in the population.

Measurement Overview

The primary functions of the measurement of staffing variables are to assess the effectiveness of the staffing function and to provide analysis to assist in compliance with laws and regulations. Based on a pre-determined set of rules measurement assigns numbers to objects to represent quantities of an attribute of the objects.

Standardization means controlling the influence of outside or extraneous factors on the scores generated by the measure, and ensuring that, as much as possible, the scores obtained are a reflection of the attribute measured. A standardized measure has three basic properties:

1. The content is identical for all objects measured

2. The administration of the measure is identical for all objects.

3. The rules for assigning numbers are clearly specified and agreed upon before the test is administered.

Scales of Measurement [nominal, ordinal, interval, and ratio]

Nominal scales classify by categories, which are given names. (The only appropriate measure of central tendency is the mode.)

Ordinal scales rank-order data, such as high, medium, low or first, second, third, but the difference between the ranks is unknown. Olympic games rankings are ordinal scales. (Appropriate measures of central tendency are mode and median.)

Interval scales rank-order data, such as number one, two, three, and there is a known, equal difference between the ranks, but there is no absolute zero. Examples include intelligence and personality. (Appropriate measures of central tendency are mean, mode, and median.)

Ratio scales are like interval scales in that there is a known, equal distance between scale points and there is an absolute zero; because of this, how much of the attribute each object possesses can be stated in absolute terms. Examples include such actions as measuring length, counting numbers or weighing items. (Appropriate measures of central tendency are mean, mode, median.)

Scores - [Scores are numerical indicators of the attributes being tested.]

Central tendency

Mean - average

Mode - most frequently occurring

Median - the center or middle value

Variability

Range - difference between high and low

Standard deviation - the average amount of deviation of individual scores from the average score; it summarizes the amount of spread in the scores. (The larger the standard deviation, the greater the variability, or spread, in the data.)

Percentile score for an individual is the percentage of people scoring below the individual in a distribution of scores (e.g., 90^th percentile indicates that 90 percent of the individuals tested scored less than this person).

Standard (z) scores represent a correction for the amount of variability in a score distribution to accurately present how well a person scored relative to the mean. A z score of 2.0 indicates that the individual received a score two standard deviations above the mean. The formula for the standard score is:

z = x - x (individual score minus the mean divided by the standard

S deviation)

Correlation indicates the strength of the relationship between two variables.

· Correlation ranges between 1.00 and -1.00 (equally powerful indicators).

· A correlation coefficient of zero shows no correlation.

· The larger the correlation coefficient, the greater the practical significance.

· If squared correlation coefficient between X and Y is .90, there is a 90% common variance shared between the two variables.

· The proper test to determine that a given sample correlation is statistically significant as an estimate of a correlation in a population is the t test.

· The smaller the level of statistical significance, the more confidence there is in the result. (A .01 level of statistical significance would provide the most confidence [99% sure] that a sample correlation coefficient would not be interpreted as having a relationship in the population, when, in fact, there is no such relationship.)

Reliability - the consistency of the results produced by a test

· Reliability studies include: test-retest; coefficient alpha; parallel forms; inter-rater.

· Coefficient alpha assesses reliability within a single time period.

· Reliability of a measure places an upper limit on the validity of the measure.

Validity - the degree to which the measure tests what it is intended to test

Validity studies include content, criterion (criterion-related) and construct. (Content validation is most appropriate when the sample size is too small for criterion validity calculations.)

· When predictor and criterion scores have been obtained, the predictor can be considered valid if the correlation coefficient has the desired practical and statistical significance.

· The case for validity generalization across situations becomes stronger if the standard error of measurement is large.

o Validity generalization is more convenient, less costly and provides more latitude in studying and using validation data.,

o Meta-analysis, which focuses on determining the average correlations between X and Y is a useful form of validity generalization.

Error – unwanted and/or missed elements in the measurement and/or measurement process

Deficiency error would indicate a failure to measure some portion of the attribute of interest; adequately define the attribute of interest, or construct a proper measure of the totality of the attribute.

Contamination error represents unwanted sources of influence on a measure.

v Standard (true) score equals Z score, (See notes, p.2.)

v True score equals the score achieved (without correction).

v Actual score equals true score plus error (actually recorded and used in decisions)

v Standard error of the measurement allows calculation of confidence intervals for true scores.