Friday, January 15, 2010
During every step in the analysis process, the analyst should observe the highest possible quality control procedures. This means that back-up copies of all data sets be maintained to facilitate recovery. Create checkpoints during the analysis process to ensure it is proceeding in the correct and desired direction. Document what actions were taken and at what point in the analysis process.
At this point the salary survey data has been collected, reviewed, arranged, aged, and warehoused and is ready for an initial analysis. Again there are numerous commercial software tools available from software providers and consulting companies to assist in this initial data preparation process. It is also possible to use spreadsheet and database tools available in the suites of various office software products to accomplish the same task, including free open source downloadable tools from the Internet.
Depending on the number and type of salary structures maintained by the organization, the analysis process begins by disaggregating the data into smaller sub-groups. These sub-groups may represent positions by line of business, geographical locations, FLSA classifications, union, non-union, plant, office, exertive, managerial, non exertive or other business unit designations.
Once the data has been sub-divided into the desired smaller groups, the analyst should review the data at the position level to determine and possibly remove any outliers. Outliers are salary survey data points that are too low or too high to be considered acceptable for inclusion in the final analysis. One method to assist in identifying outliers is to eliminate any results less than or greater than the 25th and 75th percentiles of the data set. While this is certainly a somewhat arbitrary approach, it does leave the middle 50% of results in the data set. There are more sophisticated methods available to the analyst who feels comfortable with using them including Ordinary Least Squares Regression, and Moving/Running Averages, to name but two. Most spreadsheets have built-in functions or tools to make the necessary calculations.
The analyst now has the data sub-divided into the desired data sets and cleared of any outliers. Depending on what summary statistics were reported by the survey provider or the included in a customized survey the analyst may or may not have the choice of several “averages” to use. Typically survey summary statistics may include weighted and un-weighted values for: Mean, Median, and Mode; as well as other summary statistics, such as percentiles and quartiles. Weighted summary statistics, generally, allows the analyst to have a higher degree of confidence in the reliability of the data reported.
If the salary results include salary grade designation or job evaluation points, one common and simple approach is to sort the data points for each sub-group into ascending order (Low to High) and graph the data points against the grades or evaluation points of incumbents to visually observe how they cluster or if they do cluster. By observing the graph of this relationship, the analyst will be able to conclude if the data is linear or curvilinear. Assuming the data relationship is linear, it usually is, the next step in the process is to “regress” survey Median or Mean against either the positions internal job evaluation points or their grades using Ordinary Least Squares Regression.
Ordinary Least Squares Regression is a common tool used to map the relationship between, in this case, two data variables. One data variable is the internal job grades/evaluation points and the other is the survey Median or Mean, weighted/non-weighted, as available. One output of regression analysis is a simple formula (y = mx + b) that tells the analyst the “market value” of each pay grade or job evaluation point. Most spreadsheet tools will have built in functions to calculate the various terms in the above formula.
During every step in the analysis process, the analyst should observe the highest possible quality control procedures. This means that back-up copies of all data sets be maintained to facilitate recovery. Create checkpoints during the analysis process to ensure it is proceeding in the correct and desired direction. Document what actions were taken and at what point in the analysis process.
At this point the salary survey data has been collected, reviewed, arranged, aged, and warehoused and is ready for an initial analysis. Again there are numerous commercial software tools available from software providers and consulting companies to assist in this initial data preparation process. It is also possible to use spreadsheet and database tools available in the suites of various office software products to accomplish the same task, including free open source downloadable tools from the Internet.
Depending on the number and type of salary structures maintained by the organization, the analysis process begins by disaggregating the data into smaller sub-groups. These sub-groups may represent positions by line of business, geographical locations, FLSA classifications, union, non-union, plant, office, exertive, managerial, non exertive or other business unit designations.
Once the data has been sub-divided into the desired smaller groups, the analyst should review the data at the position level to determine and possibly remove any outliers. Outliers are salary survey data points that are too low or too high to be considered acceptable for inclusion in the final analysis. One method to assist in identifying outliers is to eliminate any results less than or greater than the 25th and 75th percentiles of the data set. While this is certainly a somewhat arbitrary approach, it does leave the middle 50% of results in the data set. There are more sophisticated methods available to the analyst who feels comfortable with using them including Ordinary Least Squares Regression, and Moving/Running Averages, to name but two. Most spreadsheets have built-in functions or tools to make the necessary calculations.
The analyst now has the data sub-divided into the desired data sets and cleared of any outliers. Depending on what summary statistics were reported by the survey provider or the included in a customized survey the analyst may or may not have the choice of several “averages” to use. Typically survey summary statistics may include weighted and un-weighted values for: Mean, Median, and Mode; as well as other summary statistics, such as percentiles and quartiles. Weighted summary statistics, generally, allows the analyst to have a higher degree of confidence in the reliability of the data reported.
If the salary results include salary grade designation or job evaluation points, one common and simple approach is to sort the data points for each sub-group into ascending order (Low to High) and graph the data points against the grades or evaluation points of incumbents to visually observe how they cluster or if they do cluster. By observing the graph of this relationship, the analyst will be able to conclude if the data is linear or curvilinear. Assuming the data relationship is linear, it usually is, the next step in the process is to “regress” survey Median or Mean against either the positions internal job evaluation points or their grades using Ordinary Least Squares Regression.
Ordinary Least Squares Regression is a common tool used to map the relationship between, in this case, two data variables. One data variable is the internal job grades/evaluation points and the other is the survey Median or Mean, weighted/non-weighted, as available. One output of regression analysis is a simple formula (y = mx + b) that tells the analyst the “market value” of each pay grade or job evaluation point. Most spreadsheet tools will have built in functions to calculate the various terms in the above formula.
No comments:
Post a Comment