food insecurity in america: a macro and micro-level analysis
Post on 22-Jan-2018
45 Views
Preview:
TRANSCRIPT
1
Food Insecurity in America: A Macro and Micro-Level
Analysis
Virag Mody, Marielle Lenowitz, Aneesha Chowdhary, Eboni Freeman, Jasmyn Mackell and Ben Gross
April 12, 2017
2
TABLE OF CONTENTS
1. Executive Summary……………………………………………………………………………………..3 2. Project Goal…………………………………………………………………………………………………4
3. Part I: A Macro Perspective on Food Insecurity…………………………………………….4
3.1 Food Insecurity in the United States……………………………………………4
3.2 Selection of Exponential Smoothing and Alpha Coefficient………….4
3.3 Exponential Smoothing Data Analysis………………………………………….5
3.4 Forecasting Food Insecurity with Regression Analysis………………….6
3.5 Comparing Forecasts – Exponential Soothing vs. Regression………..7 3.6 Note on Regression Analysis………………………………………………………..7
4. Part II: Quality of Inventory at a Local Food Pantry…………………………………………7
4.1 Background on Toco Hills Community Alliance………………………......7 4.2 Data Collection……………………………………………………………………………8 4.3 P-‐Bar Chart Construction and Analysis…………………………………….….8
4.5 R-‐Chart Construction and Analysis …………………………………………….10
5. Recommendation………………………………………………………………………………………..11
6. Future Considerations………………………………………………………………………………..11
7. Appendix A (for Part II data and graphs)……………………………………………………..13
8. Appendix B (for Part I data and graphs)………………………………………………………14
9. Sources……………………………………………………………………………………………………….19
3
1. Executive Summary There is a food insecurity epidemic in America. In 2016 over 17 million American households were-‐ at some point-‐ food insecure (USDA Food Security Study). The taxpayer burden of this insecurity is massive; in fiscal year 2015, the federal government spent over 75 billion dollars on supplemental food programs (Center for Budget and Policy Priorities). While some of the hunger burden is relieved through specific federal programs, such as the free and reduced lunch program for students, SNAP, and WIC, a significant amount of food is distributed through non-‐profit entities such as food banks and food pantries. In fact, 1 of out every 7 US families at least partially relied on a food bank or food pantry in the last year to meet their needs (Feeding America Study). For our study, we first wanted to focus on a local food pantry where we could offer a recommendation that could benefit the thousands of families whom it serves each month. We chose Toco Hills Community Alliance in Druid Hills to conduct our survey, due to its proximity to Emory and the wide range of food products it receives each month. By visiting the food pantry, we were able to collect both qualitative and quantitative observations about its inventory management system and the clientele it serves. We took 16 samples of Toco Hill Community Alliance’s inventory in an attempt to calculate the approximate number of goods that are defective (expired). Using this data, we were then able to calculate that number of defective goods per million, as well as construct a P-‐Bar Chart for the number of expired goods present in each sample. Additionally, we created an R-‐Chart of the sample ranges to better understand the extent of quality management situation. From both of these charts, we found that the number of expired goods, as well as the range in expiration for expired goods, varied widely. Such variation showed that the food bank most likely does not have a system in place to ensure that goods expiring soonest are distributed first. To better understand our data, we also examined broader food insecurity trends in the US. Using data from USDA studies on food insecurity in the US from 1998-‐2015, we conducted exponential smoothing forecasts of total US households and US households with general food insecurity (as well as for subsections with low food security and very low food security). These forecasts turned out to be consistent with actual data from the period. We also ran regressions for these four categories as well, which had a noticeably higher error when compared with real data from the period. We then used the regression models to forecast the the number of food insecure households the next 5 years. Based on our analysis of the Toco Hills Community Alliance data, as well as the P-‐Bar and R-‐Charts we constructed, we recommend that the food pantry create a system that organizes goods by expiration date. Goods that are expiring sooner should be placed in the front of the room and on the outermost edge of the shelves, as a means of encouraging shoppers to pick those goods. Meanwhile, goods that are received and have several years before their expiration should be placed towards the back, because they have a significantly longer “use-‐by” date. This organizational solution will not completely eliminate the food pantry’s problem with expired goods; indeed, some of the goods the food pantry receives are already close to, if not past, their expiration date. The implementation of
4
such a system could make an impact in reducing the total number of expired goods that the pantry keeps in its inventory. 2. Project Goal This project aims to analyze the quality of inventory at the Toco Hills Community Alliance, a local food pantry in Atlanta, Georgia. Additionally, to understand the large number of food insecure households that frequent these types of food pantries, this project also sought to forecast macro level data about food insecurity, such as the total number of food insecure households each year, from 1998-‐2015 (as well as subsections of this data, such as those with very low food security). 3. Part I: A Macro Perspective on Food Insecurity 3.1 Food Insecurity in the United States The United States Department of Agriculture (USDA) defines food insecurity as a state in which “consistent access to adequate food is limited by a lack of money and other resources at times during the year.” Food insecurity exists whenever the availability of healthy, nutritionally adequate, and safe foods is limited, or the ability to obtain sufficient foods in a legitimate and socially acceptable way is uncertain. An estimated 1 in 7 Americans struggles with food insecurity. We were interested in the relationship between food pantries and food insecurity, but before we could focus specifically on our local food pantry, the Toco Hills Community Alliance, we wanted to understand the larger food insecurity problem in the United States. Using data from the USDA report entitled “Household Food Security in the United States in 2015,” we chose to forecast Total Food Insecurity as a function of Total Households, which could then further be broken down into Low Food Security and Very Low Food Security (Exhibit 8, Appendix B). By analyzing this data, we would be able to get a macro-‐level perspective on a topic that affects people both locally within the Atlanta area, as well as nationally. 3.2 Selection of Exponential Smoothing and Alpha Coefficient To properly forecast, the first step is to identify which method of forecasting is most appropriate to use. The five methods available are Naïve, Moving Average, Weighted Moving Average, Exponential Smoothing, and Regression. The following shows our analysis and applicability of each method, except for regression analysis, which is mentioned later:
• Naïve Forecasting – This method does not appropriately account for historical data, with the exception of the previous period. At a minimum, the population tends to grow positively, so using the prior period’s data point would be empirically wrong, thus eliminating this method as a viable option. • Moving Average – This method weights each data point equally, meaning that data from 1998 is just as relevant as data from 2014. Weighting older data equally to recent data would be problematic for this project because numerous factors influence levels of food security over time, such as economic trends, immigration, population changes, and health. These factors cause food insecurity to evolve over time, meaning that more current factors are more relevant to present food insecurity trends. Therefore, the historical data from 1998 should not have as much weight as recent years, removing the Moving Average as an option.
5
• Weighted Moving Average – WMA could have some applicability, but without knowing how to weight historical data, doing so would be arbitrary. This eliminates WMA. • Exponential Smoothing – This forecasting method assigns exponentially decreasing weights as the observations get older, allowing us to put more weight on more recent and more relevant data, which was the concern pointed out in the Moving Average model. This means that Exponential Smoothing is a viable method for forecasting our data.
Given that there are macro factors for variability in food insecurity, including immigration, population changes, health, and economic factors, we cannot solely rely on historical data, as there is most likely not a consistent, holistic trend. However, we cannot assume an alpha of 1 because it will become naive forecasting. Additionally, immigration, population changes, and the economy often follow trends and cycles, so to some extent, historical data is useful. Thus, to use only the previous years would be inaccurate and naïve, while discounting historical data altogether would make for a poor forecast. In order to appease both sides of this narrative, we selected an alpha value of 0.5 as a median between discounting historical data and accounting for historical information. 3.3 Data Analysis – Exponential Smoothing After forecasting using exponential smoothing, the following graphs show noteworthy information. The raw data can be found in Exhibit 1 and 2 under Appendix B. Exhibit 2 also shows the MAPE to calculate the error.
• Total Households – Our forecast for this metric is fairly accurate in tracking Historical Data, with a MAPE of 1.99%. However, except for 1998, forecasted Total Households is consistently below the actual data. This is most likely because there were variable jumps in the number of real total households, which could not be accurately accounted for, due to the fact that our exponential smoothing model weights the previous year’s forecast as heavily as the actual data. Thus, any lag in the forecast would permanently influence future predictions.
6
• Total Food Insecurity – Analysis of Total Food Insecurity be can be broken up into “Pre 2007” and “Post 2007.” • Pre 2007 – The exponential smoothing forecasts had a low forecast error because they
normalized the variability in total food insecurity. The dip from 1998 to 2000 is offset by the increase in food insecurity from 2000 to 2004. Because the model accounts for historical data at an exponentially decaying rate, the variability over time will be smoothed in our forecasted graph.
• Post-‐2007 – The massive jump in Total Food Insecurity likely resulted from the housing market collapse and subsequent recession. Our forecast model didn’t intersect the actual data from 2007 to 2013 due to our use of a 0.5 alpha. An alpha of 1 would have better accounted for the spike.
• Low Food Security and Very Low Food Security – These graphs, found under Exhibits 3 and 4 in Appendix B, provide a very similar analysis to that of the Total Food Insecurity graph. A notable difference can be seen in the Very Low Food Security Graph, whose forecast lags from 2000 to 2014. This lag is due to the same reason cited as Total Households; Very Low Food Security has been steadily increasing for years, and our exponential smoothing model has lagged as it continually accounted for historical data at an exponentially decreasing rate. Exponential smoothing limited our ability to forecast into the future to only one year ahead, 2016. If we wanted to forecast further into the future, we would have to use a regression analysis.
3.4 Forecasting Food Insecurity with Regression Analysis We used regression analysis because this method allows for forecasting beyond a single year, unlike Exponential Smoothing. Additionally, regression analysis predicts linear trends more accurately than exponential smoothing. The regression model used the same data as exponential smoothing (data which can be found in Exhibit 1, Appendix B). In analyzing the regression results, P-‐values for all different regressions are less than 0.05, which indicates significance. We thus felt comfortable using the regression analysis to forecast. Additionally, looking at the R2 values:
7
• The high R-‐Square value of 98 percent for the “Total Households” regression indicates that the regression is representative, though there may be concerns of overfitting data, which may account for noise that could impede future projections. • The R-‐Square values of the regressions for Total Food Insecurity, Low Food Security, and Very Low Food Security ranged between 64 percent and 84 percent, which indicates that there is a higher amount of variability in the actual data relative to that of our regression. (Raw numbers for p-‐values and R-‐squared are shown in Exhibit 5, Appendix B)
3.5 Comparing Forecasts – Exponential Soothing vs. Regression (Graphs of regression analysis can be found in Exhibit 6, Appendix B) Exponential smoothing is limited in how far into the future we can forecast data, but it excels at its ability to fit actual data closely. This is shown by the differences in MAPE for the comparative models. MAPE for the regression models is higher for nonlinear trends than it is for linear trends. Indeed, the only linear trend that we found was for the regression for total households. The MAPE calculations can be seen in Exhibit 7, Appendix B. MAPE for total households is much lower when the regression model is used than when the exponential smoothing model is. For total households, there are more predictable causal reasons for a linear trend. Ultimately, it is the least squares component of regression that does a better job of accounting for causal factors of change in the number of total households. 3.6 Note on Regression Analysis The regression model, while applicable for periods in which there is historical data following a linear trend, has future forecasts for years 2016-‐2020 that are likely inaccurate (forecasts for those years can be found in Exhibit 7, Appendix B). This is due to the fact that the regression model only looks at aggregate numbers and doesn’t account for causal factors. A multivariable, non-‐linear regression model would have been a more appropriate way to forecast, but we didn’t have the capability to do that for this analysis. Now that we have analyzed overall food insecurity in the United States, we can address the issues faced by our one of Atlanta’s own food pantries, Toco Hills Community Alliance. 4. Part II: Quality of Inventory at a Local Food Pantry 4.1 Background on Toco Hills Community Alliance A food pantry is defined as a charitable organization that provides those in need with food and grocery products for use and consumption at home. The food pantry we analyzed, Toco Hills Community Alliance, is a food pantry that serves DeKalb County and several of the zip codes in the surrounding area. According to its website, Toco Hills Community Alliance’s chief goal is “to provide assistance and support for individuals and families… who face the possibility of the loss of housing and/or who are without sufficient food for themselves of their families” (Toco Hills Community Alliance Website). The pantry receives a wide variety of food donations from both local grocery stores and individuals in the community. These goods are then organized into different rooms, based on the type of food item, by the employees at the food pantry. For example, one room consists of mainly canned goods and breads, while another room contains mostly snacks.
8
The food pantry follows a specific routine when serving its patrons. Individuals enter the building that houses the pantry and must prove that they qualify for assistance. Next, they are placed on a waiting list and provided with forms to complete. One by one, Toco Hills Community Alliance workers guide these individuals through the different food storage rooms. Qualifying individuals are allowed to select the types of items they want, but only workers can physically collect the groceries. At the end of the shopping period, the workers weigh the selected groceries and record the amount. Following our initial visit to the Toco Hills Community Alliance, we decided to focus on the “quality” of the inventory. For our purposes, a poor quality food item is one that is past its expiration date. We chose this aspect for analysis because the pantry’s primary goal is providing food to those in need, and thus it is important that it is serving quality food that won’t make people sick. Since Toco Hills Community Alliance does not collect information on the donations they receive, we had to use a heuristic that would represent the quality of inventory. We ultimately decided on the expiration date heuristic. By collecting expiration date data, we hoped to determine whether a quality issue existed and to give a possible recommendation to address this problem, if this turned out to be the case. 4.2 Data Collection To analyze the quality of the inventory and tracking system at the Toco Hills Community Alliance, we visited the food pantry to collect samples. We took three samples from each of the food bank’s five storage rooms, for a total of 15 samples. Each sample was obtained randomly and contained a mix of 10 perishable and non-‐perishable items. For every sample, we recorded the number of defective (expired) goods found amongst the ten items surveyed. The expiration date of an item was recorded if the item was found to be defective. See Exhibit 1 in Appendix A for the raw sample data. By taking an average of the 15 samples, we found that 34% of the sample goods were defective. This finding indicates that, on average, 3.4 out of every 10 goods at the Toco Hills Community Alliance should be expired. Converting this number to defective goods per million, we can expect that 340,000 out of every million goods donated to Toco Hills Community Alliance will be defective. 4.3 P-‐Bar Chart Construction and Analysis After collecting our data and calculating the average number of defective goods per million at the food bank, we constructed a P-‐Bar Chart. We created a P-‐Bar Chart because it can be an efficient tool to analyze the number of defective goods relative to the UCL and LCL, as well as show whether a process is out of control or not. In our case, we wanted to see the variation in defective goods among the five sample rooms and determine whether any specific rooms fell significantly outside of the average. To begin the construction of the P-‐Bar Chart, we used P-‐Bar, previously found to be 0.34, and the parameters of three sigmas, to calculate the Upper Control Limit (UCL) and the Lower Control Limit (LCL) of the data. The UCL and LCL were found to be 0.45603 and 0.22396, respectively. It is important to note that we are not analyzing a machine or production process; rather, in our case, the UCL and LCL serve as lower and upper bounds to assess if our data goes beyond these numbers when
9
analyzing the quality of the inventory. After the calculation of these values, we were then able to construct the P-‐Bar Chart. See Exhibit 2, Appendix B for the full P-‐Bar Chart calculations. Looking at our P-‐Bar Chart, represented below, we can see that the data varies widely in respect to P-‐Bar, UCL, and LCL. There are two key reasons for this vast amount of variation. First, each sample corresponds to a particular room, and some rooms contained significantly more expired goods due to the types of items that they stored. For example, Room 4 (samples 7, 8, and 9) stores goods that have a relatively short shelf life like bread. In comparison, Room 3 (samples 4, 5, and 6) mostly stores items with extended shelf-‐lives such as canned soups. Second, it was not uncommon to find a group of cans several years expired sitting next to a loaf of bread that was set to expire in a few days, when we conducted our survey. These two factors created significant variation in the data.
While the data fluctuates significantly, it is important to point out samples that fall either considerably below the LCL or considerably above the UCL. One sample that fell significantly below the LCL was sample 4, which had no defects. Two samples that significantly exceeded the UCL were samples 12 and 15, each of which had six defects. Such outliers may be due to random sampling chance, given the fact that on average, about 3.4 out of every 10 goods at Toco Hills Community are expected to be defective. It is also possible that these values are partially due to the rooms where the sample was taken, as discussed earlier. For instance, when compared with the other two samples from the refrigeration room, samples 13 and 14, sample 15 does not stand out as an outlier. 4.4 R-‐Chart Construction and Analysis In addition to making a P-‐Bar Chart, we also created an R-‐Chart. We decided to make an R-‐Chart because we wanted to analyze the range of the defective goods-‐–how long the goods in each sample
10
had been expired, relative to the day that we took the sample (April 5, 2017). Ideally, the range would be more accurate if we had information on when the item was donated to the pantry; after all, some goods may have already been expired when donated. However, since Toco Hills did not collect this information, we decided that we could best estimate this figure by comparing expiration dates to the date we took the samples. To calculate the range for each sample, we found the good with the most recent expiration date, and subtracted it from the good with the oldest expiration date. Next, we found the average of the 15 sample ranges, or R-‐Bar, which we calculated to be 10.357 months. As with the P-‐Bar Chart, we found the UCL and LCL, which were 18.3457 months and 2.2785 months, respectively. These control limits were determined using the D4 and D3 values on page 185 of the Bus351 Textbook. Calculations for the R-‐Chart can be seen in Exhibit 3 of Appendix A. The R-‐Chart, shown below for the Toco Hills Community Alliance, shows data that appears to have no distinct pattern, except a few samples (samples 13, 14, and 15). Some samples had a range of 0 months (significantly below the LCL), which would indicate that all of the defective goods in the sample had the same expiration date. Such an R value makes sense for samples 13, 14, and 15 because these samples were from the refrigeration room, where items are likely to have a short-‐term shelf life, and are consequently likely to have expiration dates close to one another. Meanwhile, some samples had an enormous range, such as samples 7 and 9, which were significantly above the UCL and had ranges of 42 and 41 months, respectively. The significant variation among R values, as well as the presence of some incredibly high values (R=41, R=42), indicates that the food bank does not have a way to monitor the expiration of goods, therefore the data suggests the need for some type of organizational system to ensure that the food pantry serves customers items that have not yet expired.
11
4. Recommendation Our analysis using the P-‐Bar Chart and R-‐Chart demonstrates that the Toco Hills Community Alliance needs an organization schedule by expiration date. To address this issue, we suggest that the pantry implement a First-‐In First-‐Out (FIFO) system to prevent donated items from reaching their expiration date while in storage. Under our proposed system, goods would continue to be organized by food type, but they would also be arranged by expiration date. For example, if a bag of apples is donated, the item would not only be placed in a room with similar items, but would also be placed near items which had a similar expiration date. Items that are close to their expiration date would in the front of the room, while items that have a longer time before expiration would be placed towards the back of the room. This layout would encourage shoppers to choose items that are close to their expiration date because those items would be in their direct line of sight when entering the room. This model mimicks how grocery stores stock their shelves. We believe the total percentage of expired goods at the Toco Hills Community Alliance would decrease under this proposal, as goods that are close to expiration will exit the pantry sooner. 6. Future Considerations While we believe that our recommendation will reduce the amount of expired goods at the Toco Hills Community Alliance at a given time, we do not believe that the inventory quality problem can be completely resolved by implementing this recommendation. This is due to the complicated reasons why the food pantry has expired goods in the first place. For example, a large portion of the pantry’s food donations come from major grocery stores in the surrounding area. These stores, however, primarily donate items that are either close to their expiration date or are already past it. This raises the issue of whether Toco Hills Community Alliance and other similar institutions should dispose of items once they expire. Such a policy would eliminate the pantry’s food quality problem – goods simply would not remain in storage past their expiration date. Many might find this solution to be wasteful and impractical. The disposal of expired items might be a net negative, as it would reduce the amount of food available. Also, some opponents of the disposal method might argue that food products are often “good” well past their expiration date, and that eating them would not cause serious illness. For these reasons, we ultimately refrained from implementing a disposal policy for expired goods. Before reaching our current recommendation, we considered the idea of implementing a spreadsheet system to record every item the pantry received as a donation. The proposed spreadsheet would record an item’s date of arrival, food type, storage room, location in the storage room, expiration date, and date of exit. After further consideration, however, we felt that this suggestion was impractical given the human capital available of the Toco Hills Community Alliance. Indeed, the pantry is relatively small and run by a few volunteers, and such a solution might prove to be too time-‐consuming. Thus, instead of recommending a spreadsheet, we decided to recommend a new organization system, a relatively simple change that we feel is better-‐suited to the Toco Hills Community Alliance’s current capabilities. However, if the pantry increases in size or gains more full-‐time volunteers we suggest that the pantry consider the idea of a spreadsheet system.
12
Lastly, it is important to mention that we had previously planned to analyze the effect that President Trump’s proposed budget cuts might have on increasing food insecurity in the United States. However, after research, it became apparent that government programs such as the Supplemental Nutrition Assistance Program (SNAP), which helps millions of low-‐income individuals in the US afford groceries, would likely be unaffected by these broad budget cuts.
14
8. Appendix B Exhibit 1 – Actual Historical Data (Household Food Security in the United States in 2015)
Exhibit 2 – Forecasting Historical Data using Exponential Smoothing
19
9. Works Cited
Coleman-‐Johnson, Alisha, Matthew P. Rabbitt, Christian A. Gregory, and Anita Singh. Household Food
Security in the United States in 2015. Rep. United States Department of Agriculture, Sept.
2016. Web. 25 Mar. 2017.
Echevarria, Samuel, Robert Santos, Emily Engelhard, Elaine Waxman, and Theresa Del Vecchio. Food
Banks: Hunger's New Staple. Rep. Feeding America, 2011. Web. 20 Mar. 2017.
"Our Organization." Toco Hills Community Alliance. Web. 25 Mar. 2017.
"Policy Basics: Introduction to the Supplemental Nutrition Assistance Program (SNAP)." Center on
Budget and Policy Priorities, 18 Aug. 2016. Web. 1 Apr. 2017.
"Supplemental Nutrition Assistance Program (SNAP)." Food and Nutrition Service. United States
Department of Agriculture, 30 Jan. 2017. Web. 25 Mar. 2017.
Wunderlich, Gooloo S., and Janet L. Norwood. "Chapter 3." Food Insecurity and Hunger in the United
States: An Assessment of the Measure. Washington, D.C.: National Academies, 2006. 41-‐54.
Print.
top related