BACKGROUND

Chapter XVI - Data Mining for Human Resource Information Systems
Data Mining: Opportunities and Challenges
by John Wang (ed)
Idea Group Publishing 2003

	Brought to you by Team-Fly

Building a Competitive Advantage

The resource-based view of the firm posits that organizational resources and capabilities that are simultaneously rare, valuable, non-substitutable, and imperfectly imitable form the basis for a firm's sustainable competitive advantage (Barney,1991). If a firm is able to uniquely manage its most valuable asset its people then it may be able to differentiate itself from the competition to create an advantage. Boxall (1993) suggests HR management can create a sustained competitive advantage by "hiring and developing talented staff and synergizing their contributions within the resource bundle of the firm (p. 66)." Huselid, Jackson, and Schuler (1997) found investments in a firm's human resources are a potential source of competitive advantage in their examination of HR effectiveness. However, many organizations lack the management know-how or the information necessary to make strategic human resource decisions.

The competitive advantage is sustained if other firms are unable to imitate their resources. The management of a firm's human resources is often less susceptible to imitation because competitors rarely have access to a firm's HR management practices, as these practices are not transparent to those outside of the company (Pfeffer, 1994). For this reason, effective HR management provides a unique opportunity for firms to establish a sustainable competitive position. One step in the process of developing a firm's human resources involves planning. Human resource planning goes hand in hand with an organization's strategic planning. As a firm develops its business strategy, it must consider the abilities of its staff to meet its strategy. As a firm identifies its strategy, HR must then ensure the strategy's success by ensuring the right HR resources are in place. HR must collect, store, and analyze data to move forward in the planning process.

All organizations collect data about their employees. However, the actions taken with that data varies widely among organizations. Some organizations use the data only to perform required administrative tasks. Others transform the data into useful information for support in decision making (Kovach & Cathcart, 1999). Using HR data in decisionmaking provides a firm with the opportunity to make more informed strategic decisions. If a firm can extract useful or unique information on the behavior and potential of its people from HR data, it can contribute to the firm's strategic planning process. Another type of information a firm may seek is the evaluation of the effectiveness of its HR practices. Often a firm implements a recruiting source or training program, for example, without any clear method of evaluating the practice. Analyzing the data collected in the implementation and management of an HR practice can provide another resource for HR-related decisionmaking.

Human Resource Information Systems

Many organizations today have purchased or designed databases to collect, manage, and retrieve human resource-related data. As stated earlier, a HRIS serves two primary purposes. The first purpose is administrative in nature. In order to conduct basic human resource transactions such as payroll, benefits administration, and government reporting, organizations must collect basic personal and work-related data. The second purpose is to support organizational decisionmaking (Kovach & Cathcart., 1999) A HRIS is often an expensive investment for a firm, and the ability to use the data collected beyond only administrative issues can help justify that investment.

The HR information systems of most organizations today feature relational database systems that allow data to be stored in separate files that can be linked by common elements such as name or identification number. The relational database provides organizations with the ability to keep a virtually limitless amount of data on employees. It also allows organizations to access the data in a variety of ways. For example, a firm can retrieve data on a particular employee or it can retrieve data on a certain group of employees through conducting a search based on a specific parameter such as job classification. The development of relational databases in organizations along with advances in storage technology has resulted in organizations collecting a large amount of data on employees.

Table 1 outlines some examples of the type of data that can be captured and stored in an HRIS. Table 2 outlines some typical functional uses of the data. Every organization varies in the type of data it collects related to its employees. However, this outline will serve as an example of the kinds of data collected and typical uses of that data.

Table 1: Data collected by an HRIS

Employee Data	Organization/Job Data
Name	Pre-employment test scores
Social security number	Job title
Date of birth	Job grade
Gender	Salary
Race	Benefit selections
Marital Status	Performance appraisal ratings
Address	Promotional history
Telephone Number	Corrective action records
Emergency Contact	Attendance history
Dependent Information	Training records
Education

Table 2: Functional use of data collected by an HRIS

Functions Using Data	Regularly Generated and Ad Hoc Reports
Compensation and Benefits	Payroll runs
Health and Safety	Benefit costs
Performance Appraisal	Recruiting effectiveness
Training and Development	Supply/demand forecasting
EEO/AA	Transaction histories
Recruiting and Placement	Training completed
Labor Relations	Adverse impact analysis
Adapted from: Fisher, Schoenfeldt and Shaw (1999).

Organizations also use data stored in their HRIS to make calculations to evaluate the financial impact of their HR-related practices. HR functions use this type of information to support the continued development of HR programs and also demonstrate to managers how HR can help impact the bottom line. Some examples of these kinds of calculations are pay and benefits as a percent of operating expense, cost per hire, return on training, and turnover cost (Fitz-Enz, 1998).

While these calculations are helpful to quantify the value of some HR practices, the bottom-line impact of HR practices is not always so clear. One can evaluate the cost per hire, but does that information provide any reference to the value of that hire? Should a greater value be assessed to an employee who stays with the organization for an extended period of time? Most data analysis retrieved from HRIS does not provide an opportunity to seek out additional relationships beyond those that the system was originally designed to identify.

Data Mining

Traditional data analysis methods often involve manual work and interpretation of data that is slow, expensive, and highly subjective (Fayyad, Piatsky-Shapiro, & Smyth, 1996). For example, if an HR professional is interested in analyzing the cost of turnover, he or she might have to extract data from several different sources, such as accounting records, termination reports, and personnel hiring records. That data is then combined, reconciled, and evaluated. This process creates many opportunities for errors. As business databases have grown in size, the traditional approach has grown more impractical. Data mining is the process of extracting information from really large data sets through the use of algorithms and techniques drawn from the fields of statistics, machine learning, and database management (Feelders, Daniels, & Holsheimer, 2000). Data mining has been used successfully in many functional areas such as finance and marketing. HRIS applications in many organizations provide an as yet unexplored opportunity to apply data-mining techniques. These systems typically hold a large amount of data a requirement for data mining. While most applications provide opportunities to generate adhoc or standardized reports from specific sets of data, the relationships between the data sets are rarely explored. It is this type of relationship that data mining seeks to discover.

Data Warehouses and Datamarts

Data-mining projects have been pursued, or at least attempted, for a long time. Several years ago, one of the authors worked in a university office of institutional research and studies. This office was charged with various queries, reporting, and decision-support assignments involving both students and personnel. Quite often, data was needed over a period of years for these projects. However, data files for previous years were difficult to locate and manipulate. Also, inconsistencies of various kinds tend to occur from year to year. For example, code systems and category definitions tend to mutate as time goes by. Similar problems have been experienced in most other organizations as well. Such projects were often practically impossible.

These problems led to the creation of special databases typically relational databases that are integrated, consistent, and multidimensional, with time being the principal added dimension. When these special databases cover all of the data for an organization, they are called data warehouses. When they cover just one functional area, such as human resources, they are called datamarts. (See for example Agosta, 2000; Gray & Watson, 1998; and Mallach, 2000). A data warehouse is essentially a separate, customized repository of data for decision-support applications (Watson & Haley, 1998). They are kept separate from operational data. Data warehouses are widely regarded as a useful resource to consolidate corporate information and share it among organizational entities for the purpose of analysis and decision-making support (Subramanian, Smith, Nelson, Campbell, & Bird, 1997). This can be a costly endeavor, and the investment may not be worthwhile depending on the use an organization expects. The decision to enable data mining often necessitates at least the implementation of a datamart. Companies who have implemented enterprise resource planning (ERP) systems will often find this facility to be included with the ERP system, which itself can be quite expensive and time-consuming. However, without at least a datamart, if not a comprehensive data warehouse, data-mining projects would be doomed to laborious preliminary data collection, editing, and formatting.

Data Mining Versus Traditional Data Analysis

It is worth reviewing here the connections and differences between data mining and the more familiar activities of reports, queries, and routine statistical studies applied to a database. Queries and reports (reports are usually based on queries) are structured questions submitted to a database. For example, if an organization wanted to compare information on two different groups of employees, queries would be needed to extract the specific employees assigned to each group for further analysis. Thus, queries are typically an essential step in extracting data from the database. Data warehouses and datamarts greatly facilitate queries involving a time dimension. For example, trend analysis and forecasts require historical data that would be hard to obtain without such comprehensive databases. Such analysis might be applied to individual employee histories as part of performance analysis. Trends in statistics relating to Equal Employment Opportunity (EEO) could also be monitored in this way. In some respects, data mining can be regarded as a set of very sophisticated queries that involve sampling and statistical modeling. In any case, the additional query capabilities and uses should be factored into the decisions on implementation of a data warehouse or datamart for data-mining use.

Decision-support system applications may also be regarded as predecessors to data-mining applications. Decision-support system applications are tools that allow users to collect data from databases, analyze the data, and represent the analysis in many forms, such as graphs and reports. These systems may use statistical modeling packages to analyze data leading to information to support decisions in organizations. These systems can combine information from other functional areas within the firm, such as customer service and marketing, to help support human resource management decisions (Broderick & Boudreau, 1992). For example, a user might build a statistical model to analyze data such as time to fill service requests and time to delivery as they relate to customer satisfaction levels. Information obtained from this analysis may then be used to establish performance goals for employees. While decision-support system applications may provide useful information for decision-makers in organizations, very specific analysis plans must be established. Essentially, the user must know what relationships exist before analyzing the data. Further, several iterations of the model may have to be tested before useful information can be identified.

One of the authors of this chapter worked in a university office of institutional research and studies (OIRS). This office was responsible for certain decision-support and data-analysis studies that often involved university personnel. There was no datamart at the time. It was necessary to create special files from current databases and retain these year after year in order to do any kinds of longitudinal analyses. Special statistical studies were sometimes possible if the right data had been collected. An example was historical trends in Equal Opportunity Employment practices. Flexibility and usefulness depended on the foresight of office analysts. If they had collected the right data, then they would be able to respond to new requests from management, state and federal agencies, and so on. It that foresight had not been in effect, it was necessary to approximate or decline the information request, or to laboriously go through old files to reconstruct a usable database. While this permitted some statistical studies to be carried out, such limited access made data mining all but impossible. For specific statistical studies, one focuses on a specific set of variables and questions. For data mining, one essentially needs all the data to be accessible in a reasonably easy environment .and the specific questions may not always be known precisely.

Descriptive statistics were often reported, especially those based on the current state of the system. Thus, it was possible for the OIRS to readily answer such questions as the age distribution of current faculty, its ethnic distribution, and gender distribution. However, to answer questions about trends in these data would require retaining the corresponding data from year to year. Ordinarily, only a small number of such yearly data sets might be retained. However, to make reasonable forecasts, for instance, a much longer series of data is required. More importantly, data-mining questions aimed at explaining the relationship of such trends to other variables in the system were practically impossible. Thus, data mining seeks to go beyond the old days of ordinary statistical reports showing only what is. It aims to find out more of the how and why.

Size of Database

An important consideration in data mining is the size of the database. As we have stated, data mining is appropriate for very large databases. While an organization may collect a large volume of data on its employees, the size of the organization may affect the suitability of applying data-mining techniques to its HRIS. Smaller organizations may mine their HR data as part of a larger data-mining project. Some organizations may have an ERP system that holds data from several functional areas of the organization. For example, in 1996, the State of Mississippi launched a statewide data warehouse project called the Mississippi Executive Resource Library and Information Network (MERLIN) (Roberts, 1999). With an interest in mining HR data, this warehouse was designed to include other data in addition to employee data, such as data on finances, payroll, and capital projects. This $5 million investment created a need to go beyond the HR function in designing a data warehouse and satisfy organization-wide needs for data mining.

	Brought to you by Team-Fly