Economic+Theory


 * [[file:Final_Stream.str]]Derek, Thiruselvan and Ilangovan**

**EXECUTIVE SUMMARY**

The basic economic theory of supply and demand states that the when the minimum wage limit is increased in a state, it results in increase in unemployment. The study conducted to evaluate this theory on fast food restaurants in New Jersey and Eastern Pennsylvania is analyzed using various data mining techniques to identify any significant changes before and after the minimum wage is increased in New Jersey. There have been studies conducted on this data before and the results have been always been opposite to what the economic theory states. But, these analyses were done just using basic regression techniques. So, the goal of this project was to evaluate the dataset using modern data mining techniques and try to find out how the fast food restaurants adapted to the increase in minimum wage. The end results using cluster analysis did not give out any significant improvement to the results of the basic regression analysis. New Jersey did not show any significant change in compensation or employment even after the wage increase while the neighboring state Pennsylvania had significant changes. There was no solid reason to support this behavior in of the fast food restaurants in response to the increase in minimum wage.

The purpose of this project is to evaluate the economic impact of an increase in the minimum wage on the fast food restaurant industry. We will analyze data collected in New Jersey and Eastern Pennsylvania in 1992 through a survey of fast food restaurants before and after the New Jersey minimum wage was increased from $4.25 per hour to $5.05 per hour. The study will help us determine whether classical economic theories about increases in the minimum wage have the predicted impacts on employment, prices, or compensation. Specifically, conventional economic theory dictates that an increase in the minimum wage will negatively impact employment: fewer workers will be hired, full-time positions will be broken into part time, or be shifted into managerial so that salaries are paid instead of hourly wages. Other expectations are that overall compensation will be adjusted but not increase, so that benefits such as health care and reduced prices on food for employees will be cut, or that the increase in wages will be passed onto consumers (inflation). If conventional economic theories are correct, the impact will be seen in one or all of these three areas. However, Keynesian economic theory argues that due to the complexities between negotiations over nominal wages, classical economic theories are inaccurate and an increase in the nominal wage will not result in higher unemployment or inflation.
 * Purpose **

Our purpose in this project is to analyze if these economic theories hold and provide recommendation to the Federal and State Governments when considering future increases in the minimum wage.


 * Literature Review**

The arguments over what impact wage increases have on employment has existed since economics [|began to emerge as a field] // //. Adam Smith’s [|The Wealth of Nations] is regarded as foundational to theories about the price of labor. The fundamental premise is that wages are determined based on supply and demand, reaching equilibrium. Since supply and demand are strongly influenced by the worker productivity, changes in the nominal wage rate will only result in inflation or output gaps as the real wages can only be affected by changes in productivity. In other words, an increase in the minimum wage will cause prices to rise, the number of people employed to fall, or total compensation (wages and other benefits) to remain the same as benefits are cut proportional to wage increases. Debates over wages and the economic impact are ongoing and aggressive. In the current economic downturn, economists such as Paul Krugman argue that cuts in the minimum wage [|would not] // //reduce [|unemployment] while conservatives insist that a [|decrease in wages] // //would do the opposite. Unfortunately, elaboration over this debate would require an in depth analysis of numerous, complex economic theories that is beyond the scope of this paper. The same is proved by David Card and Alan Krueger 1, in their research done on the minimum wage increase in New Jersey in 1992. They even collected the same data from the Bureau of Labor statistics and conducted a regression analysis and found out that there was no effect in employment due to the increase in the minimum wage. They also tried the same model when the Pennsylvania minimum wage was increased in 1996 and still they were not able to find an increase in unemployment in Pennsylvania. We expect our results to show the same results regarding the change in employment in New Jersey. However, we expect to find more information on how the increase in the minimum wage has been handled by the fast food restaurants by analyzing a lot of other factors like the price of food items, cash registers open, the number of hours the store is open and other attributes that could be adjusted to compensate the change in minimum wage.

// //http://en.wikipedia.org/wiki/Labour_theory_of_value [] // //[] [] // //[] 1 Minimum Wages and Employment: A Case Study of the Fast-Food Industry in New Jersey and Pennsylvania By DAVID CARD AND ALAN B. KRUEGER

We have collected the data from a 1992 study of New Jersey fast food restaurants before and after a $0.80 increase in the minimum wage, from $4.25 an hour to $5.05 an hour. For a control group, a comparable survey of fast food restaurants was conducted at the same time. Response variables are the number of full time employees, part time employees, and managerial employees, price of the products, compensation and salary.
 * Data Understanding: **

Before starting to prepare the data, a new variable “B4/After” was included in the dataset to make the clustering possible. “B4/After” is a flag variable and it tells which row is data for before the interview and after interview. “0” implies before interview and “1” implies after interview data. This variable will be helpful in clustering the data and to identify the changes happened after the minimum wages were increased. After examining the data, we found that all the variable fall into Quantitative variable and Nominal variables as follows,
 * Data Preparation **
 * ** Quantitative ** || ** Nominal ** ||
 * NCALLS || CHAIN ||
 * EMPFT || CO_OWNED ||
 * EMPPT || STATE ||
 * NMGRS || SOUTHJ ||
 * WAGE_ST || CENTRALJ ||
 * INCTIME || NORTHJ ||
 * FIRSTINC || PA1 ||
 * BONUS || PA2 ||
 * PCTAFF || SHORE ||
 * MEALS || TYPE2 ||
 * OPEN || SPECIAL2 ||
 * HRSOPEN || STATUS2 ||
 * PSODA || B4/After ||
 * PFRY ||  ||
 * PENTREE ||  ||
 * NREGS ||  ||
 * NREGS11 ||  ||

The first step in data preparation is we filtered out TYPE2, SPECIAL2, STATUS2, DATE2 and NCALLS variables as they are just the type of communication used to collect the data and have no significance to our analysis. Based on our category we determined the type of each variable. Once the type is determined we partitioned the data into 50% training and 50% testing. Then Auto data preparation was done on the data to replace the missing values with the mean value of that variable and since the values in each variable are not in the same unit z-score transformation was done on the data. PCA factor was run in this data to see if there is any high correlation between the predicting variables which might lead to multicollinearity and reduce the accuracy of the model. Then Two Step cluster model, K-mean cluster model and Kohonen cluster model to see the performance of this models in this current dataset. Then to improve the accuracy of the model several variables (CHAIN, CO_OWNED, SOUTHJ, CENTRALJ, NORTHJ, PA1, PA2 and SHORE) were filtered out to ensure that the model is not affected by multicollinearity of these variables with STATE variable. This was done to see if New Jersey comes as a different cluster which would enhance our interpretation of the economic theory. This is the final data used to perform the model. ** Descriptive statistics of the give data set was found for the entire data set and then for New Jersey and Pennsylvania respectively to identify if there is any difference in mean, median, mode and standard deviation of the control variables. Autocluster is used in beginning of our modeling process to determine which cluster would yield a better result for the given data. Secondly, Two Step clustering model was run on the dataset. We ran two step clustering model first because in this cluster model, number of clusters will be determined by the model itself. Then K means clustering and Kohonen clustering were performed in the dataset with the same number of clusters given by two step clustering model. Three clustering models were created to compare the result with one other and to choose the best model of the three. Then select node is used to select only the New Jersey data to see if there is any significant difference in control variables before and after the interview. Logistic regression and feature selection was performed on that data to check the most important variable in determining the B4/After variable. This will help us to understand which variables were affected the most, before and after the interview were conducted. Both logistic regression and feature select was performed to compare the result with each other and ensure the accuracy of predicted variable. Also the three clustering models Two Step, K means and Kohonen were run again with the select data for New Jersey to see if before and after interview is separated as two different clusters. Another method we used is to filter the most important variables from the entire data set (WAGE_ST, INCTIME and MEALS) in predicting the B4/After variable and run the three clustering model again to see any significant change in clustering the Pennsylvania and New Jersey as two separate clusters. Again logistics regression was performed in the whole data and the next two important variables (PSODA and PFRY) were filtered and the clustering was performed to check if Pennsylvania and New Jersey were separated as two different clusters.
 * Modeling

Results
__Kohonen__ – The auto cluster function of modeler suggested Kohonen would provide the best results for cluster analysis. However, the results were poor, with a silhouette value of 0.1. The number of clusters were determined to be 4 by the Two Step model and then applied to the Kohonen. This did not give a clear separation between the before and after wage increase period. __K-means­__ – The initial results were promising. By specifying four clusters, a silhouette value of 0.3 was obtained. However, further examination of the data discovered that contrary to expectations, New Jersey, before and after the minimum wage increase, formed one cluster, and that Pennsylvania formed the other three. Theorizing that irrelevant data might be creating the unusual results in the initial analysis, a filter node was added to the model. Data regarding the chain of fast food restaurant and the location variables were reduced to a single flag variable; New Jersey or Pennsylvania. __Two Step__ – This resulted in a silhouette of 0.2 and had 3 clusters. This number of clusters was used in the other two models. __Kohonen__ – Again, the silhouette value was 0.1 although the number of clusters had been reduced to 3 __K-means__ – Using three clusters, silhouette value remained at 0.1. Still there was no difference in the New Jersey cluster. New Jersey remained in a single cluster. Same kind of analysis of using logistic regression to find the most important predictors, removing them and doing the cluster analysis turned no significant improvement to the model. New Jersey remained as a separate cluster till the end. From the logistic regression, the significant variables were starting wage, the percentage of employees affected by the minimum wage increase, and the length of time to the first bonus. The cluster analysis on just the New Jersey data showed very low silhouette values and did not consider the before and after the increase to be an important variable at all. So, again there was no significant finding showing a good difference between before and after the wage increase in New Jersey.
 * All Data – Transformed**
 * Filtered Data**
 * New Jersey Only**

Conclusion
The multiple analysis failed to provide significant evidence an economic impact caused by the New Jersey increase in the minimum wage. In fact, substantially more variation in was observed within the Pennsylvania control group than New Jersey. There were significant changes in the Pennsylvania data, before and after the wage increase in New Jersey. However, there is no statistical or qualitative information available that could explain this variation in Pennsylvania. The reason for the New Jersey cluster remaining unchanged throughout the analysis could be because of the following 1. Most restaurants had increased the minimum wage even before it was officially announced. 2. The restaurants have changed their cost structure and were ready for the increase in minimum wage well before it was officially announced. The same study was done with data from Pennsylvania after there was an increase in minimum wage in Pennsylvania by Card and Krueger and they found no difference before and after the increase. This shows evidence that fast food restaurants are prepared for the increase in minimum wage increase. The only significant difference that could be noted before and after the increase of minimum wage is using the basic descriptive statistics. It shows that the time taken for the first increment has increased by almost 3 months after the minimum wage has been increased. To conclude, the increase in minimum wage has little or no effect to the employment in New Jersey’s fast food restaurants and the basic economic theory does not hold good for fast food restaurants.

- All the variables are defined in the excel file