Basic Theories to the Conceptualizations of Job Analysis Accuracy.

There are two basic theory to the conceptualizations of job analysis accuracy:- Classical Test Theory and Generalizability Theory. Classical test theory suggests that a "true score" esists for a given job and that true scores are stavle over time. Any measurement variation error are eliminated or aggragated across time or sources. Through this, researchers have commonly aggregated job analysis information in order the get the "true score" for every particular job, then the quality of job analysis datas are indexed with estimation of interrater reliability.

However, classical test theory can only estimate one source of error at a time. To overcome this, generalizability theory which concerns with the dependability of behavioral measures are adopted. The dependability involves the accuracy of generalizing from an observed score to the average score over different kind of situations. However, this theory allows one to segment the sources into multiple sources or facets, such as the methods of data collection and sources of data. The ability to estimate multiple sources of measurement error simultaneously enable us to examine how different methods, sources of data, location, incumbent abililty and other possible factors impact the accuracy of job analysis data.

Cronbach's (1995) method of differentiating the accuracy scores have four components:- elevation, differential elevation, stereotype accruracy, and differential accuracy. Elevation refers to the way raters use response scales and is a function of the differences between the average of a person's ratings and the average expert score. Differential elevation refers to how close an average job rating across all dimensions to an average expert rating would be. The abiltiy to predict the profile of dimension means across the jobs are reflected by stereotype accruracy while diffrential accuracy reflects the abliity to predict the differences between jobs and individual dimensions.

Morgeson and Campion (1997) came up with Multidimensional Conception and suggested that inaccruracy can be categorized into six ways:- interrater reliability, interrater agreement, discriminability between jobs, dimensionality of factor structres, mean ratings, and completeness of job information.

Interrater reliability is the consistency across different raters and indexes arate covariation while interater agreement refers to the levle of agreement across different raters. Discriminability between jobs refers to the abitlity to differentiate between different jobs. The extent to which factor structures are comples or multidimensional are reflected by the dimensionality of factor structures. Mean ratings refer to elevated or depressed ratings, and finally, completeness of job information refers to the relative comprehensiveness of the job analysis data.

These six categories reflect the underlying issues of reliability and validity. It represents a multidimensional conception of accuracy and that accuracy can be indexed in many different ways. Any single estimate may faile to sufficently assess the accuracy of job analysis data. It is also possilbe that job analysis data are affected along one or two of the dimensions and that some of the indices require researchers and practitioner to identify what is reasonable in their job analysis data. Other than these, higler or lower levels of these indices could indicate inaccuracy.

All those conceptions of accuracy mentioned above rely heavily on the notion that a "true score" exsists. However, such an assumption may be problematic in the job analysis context such as more dynamatic work settings and environments. Some even suggested that the nature of tasks performed by incumbents can change dramatically over time. If jobs change over time, the true score model would not be able to be applied and that a "true score" as defined in classical test theory could not exist if jobs are socially contructed.

If accuracy is viewed ass convergence to a known standard, then the job analysis data's accuracy will be inappropriate since there are rarely unambiguous standards against which to judge these data. However, there are various ways to index convergence and it may be able to reflect accuracy. It is relatively easy to show how the datas might not be objective in an absolute sense if one focuses heavily on the accuracy of job analysis data. Therefore, the data's accuracy may never be certain. In order to avoid these difficulties, Sanchez and Levine (2000) focused on consequential validity as a standard for job analysis accuracy.

As noted by Morgeson and Hofmann (1999), focusing on outcomes has substantial precedent in other areas of organizational science but as a standard for job analysis accuracy, consequential validity has two difficulties:- it reflects usefulness more than accuracy, and the problems which are associated with making judgments about job analysis data. This eventually means that Sanchez and Levine's (2000) conception doesn't address the difficulties linked with job analysis data to evaluative standard. Because of this, attention are shifted to the consideration of the validity of job analysis inferences. This shift avoided the problems in associated to the discussion of job analysis accuracy, and implicitly includes Sanchez and Levine's (2000) notion of utility.