Decision Analysis Model and Report

JaKaiser Smith
Southern New Hampshire University
Date: 07/09/2015

Abstract
In this report following resources have been utilized to establish a relationship between Retail Salesperson’s salaries and their intent to shoplift at their own workplace:

* The Larceny theft data from Federal Bureau of Investigation’s official website for the years 2011, 2012 and 2013; * 25th and 26th Annual Retail Theft Surveys by Hayes International for the years 2011, 2012 and 2013; * National Conference of State Legislatures website for Labor and employment data for the years 2011, 2012 and 2013.

‘Shoplifting’ is undoubtedly a psychological issue for most of the people. Shoplifting for most individuals is rarely about greed or poverty. It’s about people struggling with their own personal conflicts and needs. There are approximately 27 million shoplifters (or 1 in 11 people) in the USA today. More than 10 million people have been caught shoplifting in the last five years. The pleasure produced from “getting away with it” yields a chemical reaction causing in what shoplifters describe as an incredible “rush” or “high” feeling. Many shoplifters have committed to the fact that this high is their “true reward,” rather than the stolen product itself.
It is assumed in this project that all the shoplifting instances in different states of the USA have been performed by the front-end Retail Salesperson. This might not be 100% true as the members of the Retail Store management of a company may also get themselves involved in such heinous crime, but this possibility has been disregarded. Data from the three aforementioned sources were compiled according to the respective states, years and concrete data-sets were formulated. MS Excel’s TreePlan was then used to create Decision tree models in order to determine whether the Retail Salesperson’s salary actually had an effect on their choice to shoplift at their own workplace and that how much of the increment does the current salary points needed in order to lower down the shoplifting rate i.e., shoplifting instances per 10,000 Retail salesperson.

Table 1 Research Analysis Synopsis (Year: 2013)
The Table 1 above summarizes the findings from this project. Here A_PCT10, A_PCT25, A_MEDIAN represents average 10th percentile, 25th percentile and Median salaries of Retail Salesperson for the year 2013 respectively. There is a small yet clear disparity between the salaries where shoplifting instances are less than or equal to 378 and that where the same is greater than 378. When the salaries are at the extreme low end of the spectrum even a difference of $100 makes a lot of difference in the life of an individual!

Table of Contents Introduction 4 Research Question 4 Data Appraisal 5 Techniques 6 Evaluation 7 Model 8 Results 10 Limitations 11 References 12 Annexures 13

Figure 1 Decision Tree (Scenario#1) Year: 2013 8 Figure 2 Decision Tree (Scenario#2) Year: 2013 9 Figure 3 Colour key for Decision Tree (Scenario#2) Year: 2013 9 Figure 4 Resultant values from the two Decision Trees 10 Figure 5 Sensitivity Analysis (as per A_PCT10 in Decision Tree#2) 10 Figure 6 Shoplifting rate graph 13

Table 1 Research Analysis Synopsis (Year: 2013) 2 Table 2 Core Values used for plotting Decision Trees (Year: 2013) 7 Table 3 Core Values used for plotting Decision Trees (Year: 2012) 7 Table 4 Master Data-set 14

Introduction

‘Shoplifting’ is a major crime in the USA today. According to the data sets used in this project, it has been estimated that the annual shoplifting losses in the USA amounts to be greater than $13,000,000,000! On an average an employee shoplifts products worth $715 dollars whereas the average amount shoplifted by an External customer is around $129. This is confirmed by the fact that Employee theft contributes 43% and shoplifting contributes 37% to the overall Retail shrink in the USA. The overall Retail shrink in USA is measured at around 1.48% of the total Retail sales. For example, the Retail sales in the USA for the year 2014 was around $4.732 trillion, therefore, the Retail Shrink during that year would have been $70,033,600,000! Even though technologies like Smart Tagging, Entry sensors, and Source tagging have been deployed but still the numbers don’t cease to proliferate.
This analysis is specific to Retail Salesperson, the front-end force of people who drives the sales in the stores. The intended population for this analysis is that of the Retailers who always have a hard time in figuring out the ways to minimize the store shrinkage. As the Store management team are reasonably well paid, well-educated and need to set an example for these Sale-force, the personnel responsible for store shrink have solely been considered to be Retail Salesperson (OCC Code: 41-2031). All the data pertaining to ‘Retail Salesperson’ is used in this research (e.g., number of employees and their current pay for the year 2011-13). The latest available data are that of the year 2013, therefore, the analysis has been conducted for the period of 3 years ranging from 2011-13.

Research Question

“What change in the Retail salesperson’s salary (10th percentile, 25th percentile and median amounts) is needed to keep shoplifting instances by employees to a level below average of 378 for the year 2013?”

In this particular project the states with top 30% shoplifting instances are considered as a yardstick over the period of 2011-13. TreePlan via MS Excel is used to carry out this analysis. TreePlan helps in building a decision tree diagram in an Excel worksheet using dialog boxes. Decision trees are useful for analyzing sequential decision problems under uncertainty. TreePlan creates formulas for summing cash flows to obtain outcome values and for calculating rollback values to determine optimal strategy. Since the population, number of retail employees, area and other demographic factors may contribute to the overall shoplifting trends, the typical 50% above and below averages is not considered in this project, in order to arrive at some meaningful figures.
The data from all the three years 2011-13 is combined to have 156 samples for the Salary-shoplifting analysis. The average shoplifting instances (per 10,000 Retail salespersons) are 322, 310 and 282 for the year 2013, 2012 and 2011 respectively. This clearly shows a steady increase in the number of instances. Hence, it is imperative to understand the reasons behind such increase. It was interesting to comprehend that an increase in the salary of the Retail salesperson could bring this alarming average rate down. It should be noted that the specific value of 378 as a benchmark of the shoplifting instances is calculated by taking the same value for lowest rank state amongst the group of top 30% pertaining to the year 2013. If the year 2012 is considered for the similar analysis, the threshold value comes out to be 354.
Data Appraisal

The data utilized in this project has been sourced from three main web portals. The data pertaining to the Larceny theft has been segregated from Federal Bureau of Investigation’s website. As an intelligence-driven and a threat-focused national security organization with both intelligence and law enforcement responsibilities, the mission of the FBI is to protect and defend the United States against terrorist and foreign intelligence threats, to uphold and enforce the criminal laws of the United States, and to provide leadership and criminal justice services to federal, state, municipal, and international agencies and partners. The Uniform Crime Reporting (UCR) Program has been the starting place for law enforcement executives, students of criminal justice, researchers, members of the media, and the public at large seeking information on crime in the nation. The program was conceived in 1929 by the International Association of Chiefs of Police to meet the need for reliable uniform crime statistics for the nation. In 1930, the FBI was tasked with collecting, publishing, and archiving those statistics. Today, four annual publications, Crime in the United States, National Incident-Based Reporting System, Law Enforcement Officers Killed and Assaulted, and Hate Crime Statistics are produced from data received from over 18,000 city, university/college, county, state, tribal, and federal law enforcement agencies voluntarily participating in the program. The crime data are submitted either through a state UCR Program or directly to the FBI’s UCR Program. In addition to these reports, information is available on the Law Enforcement Officers Killed and Assaulted (LEOKA) Program and the Hate Crime Statistics Program, as well as the traditional Summary Reporting System (SRS) and the National Incident-Based Reporting System (NIBRS). Over 1.2 million shoplifters and dishonest employees were apprehended in 2014 by just 25 large retailers who recovered over $225 million from these thieves, according to the 27th Annual Retail Theft Survey conducted by Jack L. Hayes International. For over 30 years, Hayes International has helped all types of retail, industrial, distribution and manufacturing companies to become a great deal more profitable through a program of effective inventory shrinkage control and risk management. Data pertaining to the average values such as Employee theft amount, percentage of employee theft contribution to the overall retail shrink, etc. have been derived from the data sources available at Jack L. Hayes International website. The number of Retail Salesperson in different locations of the USA and their respective salary points have been sourced from National Conference of State Legislatures website. In 1974, three organizations represented the interests of legislators and staff, but their influence was diluted. So seven inventive legislative leaders and two staffers got together and envisioned a single national organization to support, defend and strengthen state legislatures. After a survey of lawmakers and staff from around the country confirmed their idea was a good one, the three organizations dissolved, and on Jan. 1, 1975, the National Conference of State Legislatures was born. Labor and employment issues are covered by a combination of state and federal laws and are important to workers, businesses, labor organizations and governments. NCSL’s resources on labor and employment issues are arranged around six topic clusters: Collective Bargaining, Discrimination, Employee Leave, Personnel Issues, Unemployment, and Wage and Hour.
Techniques

There has been a growing trend in organizational and business management that involves decision-making, based upon accurate data analysis. Every organization requires reliable decision and data analysis tools. Fortunately, a spreadsheet program in Excel, which is an integral part of Microsoft Office, is greatly helpful in this regard. Excel is an incredible tool used by majority of organizations for their data analysis requirements. It has user-friendly features.
These include an array of diagrams and graphics to help commercial companies in making a wise data-driven decision. TreePlan helped in building a decision tree diagram in an Excel worksheet using dialog boxes pertaining to this project. Hence, the Decision trees produced were useful for analyzing sequential decision problems under uncertainty. TreePlan automatically included formulas for summing cash flows to obtain outcome values and for calculating rollback values for determining optimal strategy.
The trees are "solved" using formulas embedded in the spreadsheet. The terminal values sum all the partial cash flows along the path leading to that terminal node. The tree is then "rolled back" by computing expected values at event nodes and by maximizing at decision nodes; the rollback EVs appear next to each node and show the expected value at that point in the tree. The numbers in the decision nodes indicate which alternative is optimal for that decision.
A decision tree can be used as a model for a sequential decision problems under uncertainty. A decision tree describes graphically the decisions to be made, the events that may occur, and the outcomes associated with combinations of decisions and events. Probabilities are assigned to the events, and values are determined for each outcome. A major goal of the analysis is to determine the best decisions. Decision tree models include such concepts as nodes, branches, terminal values, strategy, payoff distribution, certainty equivalent, and the rollback method.
From the data combination and calculation, it can be clearly seen that the Larceny theft instances have undergone a negative Growth rates through 2011-12 and 2012-13. Remaining parameters (like Population, amount shoplifted, shoplifting instances) have seen a sluggish growth rate during this period but none have gone negative when compared to that of the previous year. If the Rates of Overall US Population is seen in contrast with the Total Retail Salesperson, it can be deduced that even though the growth rates of the former has been sluggish but that of the latter has been quite high.
This may imply that US people are willing to be a Retail Salesperson during early part of their career and later use that experience to be a part of Retail Store Management. The hypothesis that the ‘Retail Salesperson’ (OCC Code: 41-2031) were solely responsible for employee theft can be an issue here as it is highly unlikely that most of the job seekers would work as a salesperson in Retail environment with theft being their only motivation. The exact values pertaining to growth in Total amount shoplifted and Employee theft’s amount comes from the fact that the latter values were calculated by averaging out. Due to the unavailability of the data belonging exclusively to a particular state and location, the average values of the entire country was used to calculate the respective sub-values.

Evaluation

Parameter | A_PCT10 | A_PCT25 | A_MEDIAN | Overall average values | $16,885 | $18,252 | $21,146 | Considering Shoplifting instances <= 378 | 39.47% above average | 36.84% above average | 50.00% above average | Considering Shoplifting instances > 378 | 33.33% above average | 20.00% above average | 20.00% above average | Average of states when
Shoplifting instances <= 378 | $16,876 | $18,263 | $21,251 | Average of states when
Shoplifting instances >378 | $16,909 | $18,224 | $20,882 |

Table 2 Core Values used for plotting Decision Trees (Year: 2013)
The Table 2 has the core values from which the two Decision Trees for the year 2013 are made. The values in the Table has been arrived at by calculating the respective values of all the locations in the US for the year 2013. The required values were taken from the three sources mentioned earlier. Now that the data is identified, transformation and evaluation of the data can take place in a top-down model. The first step for the top-down model is gathering all the data into a single spreadsheet which can be analyzed using Excel and R-Studio. For the purposes of developing a decision tree, 30% is the break point for the data. The reason for taking 30% as the threshold value is owing to the fact that taking 50-50 average would have not served the purpose to distinguish the good locations from the bad locations as per the shoplifting rates. Top States with shoplifting rate of 378 or less are considered desirable, while states with shoplifting rates greater than or equal to 378 are considered undesirable. This 378 as value is derived by the fact that all the top 30% states have shoplifting rate of 378 or more. 71.15% of the states showed shoplifting rates of 378 or less. 28.84% of the states showed shoplifting rates greater than 378.

Parameter | A_PCT10 | A_PCT25 | A_MEDIAN | Overall average values | $16,903 | $18,286 | $21,202 | Considering Shoplifting instances <= 378 | 29.73% above average | 48.65% above average | 48.65% above average | Considering Shoplifting instances > 378 | 26.67% above average | 26.67% above average | 33.33% above average | Average of states when
Shoplifting instances <= 378 | $16,875 | $18,321 | $21,324 | Average of states when
Shoplifting instances >378 | $16,973 | $18,199 | $20,901 |

Table 3 Core Values used for plotting Decision Trees (Year: 2012)
The Table 3 can be used to construct Additional Decision Tree for the year 2012 in the similar way as it is done in this paper for the year 2013.
Model

After the values from Table 2 are imported into MS Excel Tree Plan a value of $56,390 is arrived at, this is summation of A_PCT10, A_PCT25 and A_MEDIAN giving a point in the range where the ideal summation may lie. When the average values of these 3 parameters (when Shoplifting instances >378) are added after taking the numbers from Table 2, a value of $56,015 is arrived that. This clearly shows that there is disparity which ought to be addressed to and the salaries out to increase to at least $56,390.

Figure 1 Decision Tree (Scenario#1) Year: 2013
After implementing a regular weighted mean algorithm, $17,022, $18,346 and $21,022 values are derived for A_PCT10, A_PCT25 and A_MEDIAN respectively, which show that there should be a slight increase in the salaries of the Retail Salesperson as prompted by the MS Excel Decision Tree in the Scenario#1. Similarly, for Scenario#2 as per the Figure 2, the values $17,032, $18,357 and $21,034 are obtained for the three salary points.

Figure 2 Decision Tree (Scenario#2) Year: 2013
It is important to note that the average value of A_PCT10 i.e. the 10th percentile salary is more for the states with shoplifting instances greater than 378 than the states with that less than 378 with reference to the data is Table 2. This implies that the lower values of 10th percentile salaries need not necessary persuade the Retail Salesperson to indulge in theft at their workplace. But when the values of A_PCT25 and A_MEDIAN are taken into consideration, they follow the hypothesis stated earlier that lower salaries instigate Retail Salesperson to indulge in theft in the stores. Although the difference between the salaries with the same head is not large, but it should be kept in mind that very low salaries account for low Dispensable income which can in-turn result in theft by the employees.
In the first Decision Tree (Figure 1), the decisions followed the order of A_PCT10, then A_MEDIAN and then A_PCT25 at respective nodes which resulted in the TreePlan giving the summation value of $56, 390 for the three percentile salaries. If the order of the decisions are changed to A_PCT25 at the first node, A_MEDIAN at the second node and A_PCT10 at the third node, a TreePlan seen in Figure 2 is obtained.Figure 3 Colour key for Decision Tree (Scenario#2) Year: 2013
Figure 3 Colour key for Decision Tree (Scenario#2) Year: 2013

Results

Figure 4 Resultant values from the two Decision Trees
Although the difference in the values of the same head is not large but it has to be noted that a difference of $100 makes a lot of difference in the lives of Retail Salesperson. Figure 4 shows values deduced from the two Decision Trees. A major part of their salary goes into the mere survival of themselves and their family. It can also be seen for the analysis that A_PCT25 and A_MEDIAN needs to be re-worked by the concerned authorities as there is a clear disparity in their values in all the aforementioned situations. Figure 5 is the Sensitivity Analysis graph for the 10th percentile salary point in the Decision Tree#2.

Figure 5 Sensitivity Analysis (as per A_PCT10 in Decision Tree#2)
Limitations

This project, like all projects, has limitations. As discussed in this paper, there are other factors that could be studied for their relation to shoplifting rates. This analysis is also limited by the availability and granularity of the data. This analysis is performed on the state level. If some of the data was available by city level, it would have added much more detail to the analysis, and potentially better decision making. With data mashups, the data used was not designed for the type of analysis that is being performed. This particular model uses the most granular data available for the mashup. The granularity of the model is one of its limitations.
The model results in a three values for the salary points for the year 2013. Changing one piece of data, would in turn change the associated shoplifting values and would change the overall 10th percentile, 25th percentile and Median salary values. Therefore, it is difficult to compare salary values from the three variables evaluated in this analysis to three different salary variables. This is another limitation. The model can be updated with new data of the same kind, but is changed if data is swapped for other types.
To explore further limitations of the model, alternate decision trees can also be developed and modeled, to show if the original analysis was subject to confounding and bias, both which cause errors in decision analysis.

References

https://www.census.gov/content/dam/Census/library/publications/2014/acs/acsbr13-02.pdf https://www.census.gov/prod/2013pubs/acsbr12-02.pdf http://www.ncsl.org/research/labor-and-employment/2012-state-unemployment-rates.aspx https://www.fbi.gov/stats-services/crimestats http://www.ucrdatatool.gov/Search/Crime/State/RunCrimeStatebyState.cfm http://hayesinternational.com/news/annual-retail-theft-survey/ Annexures

Figure 6 Shoplifting rate graph

Table 4 Master Data-set