Planning that lays out your step-by-step sampling strategy is absolutely essential prior to stepping foot out into the field. But, even before you work through those steps, by way of background it is useful to review the whys and wherefores of waste sorts.

This chapter [A] recaps the purpose of waste sorts and then [B] explains the steps to develop a study design.

^[A] Why do a waste sort?

Essentially, in order to plan solid waste strategies, it is necessary to first understand something about the waste stream generated by an area's population. Examples are when you need to know
 

Sizing requirements for material recovery facilities

Projected sales from recyclables

Proportion and amount of organic material for composting

Effectiveness of mandated recycling programs

Amount of potentially divertable material justifying inclusion in future recycling efforts

Changes in disposal patterns over time

Heat content of the incoming incinerator waste stream
 

Unfortunately, because waste composition varies so much from place to place and from time to time, "generic" data from other sources is essentially useless for answering these questions. The only meaningful composition data is that which you develop from your own locale.
 

Since it is obviously impractical to conduct a complete census of the entire waste stream in your area, it is necessary to estimate the characteristics of the whole by sampling a small part of the total. It is the science (and art) of statistics which makes it possible to insure that your estimates are representative and to know how close those estimates are likely to be to the correct answer. Otherwise, you will not be able to use the sample averages in a way that recognizes the possibility of errors inherent in any sample of the whole. But all this will only work if you follow the necessary steps.
 

A good study design begins with asking the right questions and then following the steps listed below. Before walking through those procedures, however, you really ought to prep yourself on the on the underlying statistics and on the physical lay of the land.
 

First, the Stat / _facts Primer  explains most of what you'll want to know about the underlying theory of statistics to construct a good study design and conduct a reliable waste sort. Most people consider a detour through statistics akin to being forced to eat cold porridge, but the rest of this GUIDE will make more sense if you do.
 

Second, to help insure that your plans conform to reality, drive around for a week to see how solid waste is handled and chat with the major players in the field. This includes following some collection vehicles on their routes, watching trucks arrive and dump at the disposal facility, and interviewing program managers, haulers and facility operators on how their systems work.
 

After doing this and perusing the Primer, you will be in a better position to understand and implement the following discussion of the eight key steps in a waste sort study design: [1] the question (p. 9); [2] the waste stream (p. 10); [3] the waste shed (p. 11); [4] any groupings by source of waste (p. 12); [5] the sampling strategy (p. 17); [6] the materials sorted (p 20); [7] time periods involved (p. 22); and [8] the desired level of reliability (p. 22).
 

[Note: If you don't want to wrestle with statistics but do want to enjoy the fruits of what statistics can provide you may want to consider purchasing the WasteSort_™ software package to automatically handle all of those calculations for you. An order form is contained on the last page of this GUIDE.]
 

[1] State the question being asked
 

To frame your study design, articulate the question(s) you are trying to answer, expressed in as clear and precise a manner as possible.
 

[example: What changes occur in the composition of waste by weight after the introduction of curbside recycling programs for metal cans, glass and plastic bottles and newspapers by the city of Middletown?]
 

[another example: If plastics are diverted from the waste stream by recycling, will the remaining heat content be sufficient to sustain economic operation of a waste-to-energy recovery facility?]
 

[2] Define the waste stream
 

Implicit in the questions you are seeking to answer is the type of solid waste in the waste stream that you need to characterize. That is to say, are you only interested in household waste, or do you want to consider distinct industrial types, such as hazardous or construction/demolition debris. One standardized breakdown of waste by type, or, for that matter, by source, is shown on the following table

* Sources are indicated by a number in brackets, the reference for which is found in the appendix: More Readings, at p. 119.
 

The first two columns are sometimes referred to as household waste, the first four columns are sometimes combined under the heading municipal solid waste, and the last is sometimes referred to as special wastes.
 

Later, if you chose to only characterize a part of the entire waste stream, you will need to isolate loads which come from just the types of waste you are analyzing. Often these different types will come on either discrete types of trucks (i.e. residential and commercial from packers and demolition debris from roll-offs) or from certain companies (i.e. Household Hauler Co. or Axle Demolition Service), making your task easier.
 

[3] Define the waste shed
 

Next, define the "waste shed," the geographic area which encompasses the source of the waste stream waste which you wish to characterize.
 

[example: Middlebury County (and the City of Middletown 35 miles away) is the waste shed for an incinerator whose operator has a flow control contract with all solid waste generators in the County and in Middletown and who wants to evaluate the heat content of the incoming waste stream.]
 

[another example: The City of Axleberg municipal boundary is the waste shed for a study to find out how effective the city's new curbside recycling program has been by comparing the proportion of recyclables in its discards before and after the program began.]
 

Note that the area does not have to be composed of contiguous parts. Especially due to the increasing use of transfer stations, waste can be economically moved long distances when it makes economic or political sense to do so. The relevant factors for drawing boundaries can be unrelated to geographic proximity. Other driving factors are
 

Economic issues (in terms of costs of long distance hauling),
 

Political issues (in terms of public acceptance of nearby disposal facilities), and
 

Legal issues (in terms of mutually agreed upon contracts or flow control ordinances to capture discards from an area).
 

[4] Categorize the source of waste generators by type
 

The fourth step involving grouping the samples by distinct sources is often done. However, it does not have to always be done. Categorizing waste by the source which generated it is usually referred to as stratification. To stratify your samples, it is necessary to be able to
 

Identify loads by the group which generates the waste; and
 

Estimate the proportion that each strata bears to the entire waste stream.

(See Table 1 on p. 10 for a list of types of solid waste that can also relate to sources of waste for stratification.)
 

Most often, residential and commercial sources are separated into strata. This division is relatively easy to make because the two have distinct forms of collection residential waste is collected from individual 30 gallon cans set out at the curb and commercial waste, in 2-6 cubic yard dumpsters. For this reason, the two are generally collected in separate trucks with discrete loads. When construction and demolition debris and other high volume wastes are important to characterize, they may be split off into yet another strata. This is usually not difficult to accomplish, because they are often collected from roll-off or other specialized containers.
 

Stratification affords three benefits: [a] it provides discrete information about each strata that you need to answer your questions; [b] it reduces the number of samples you need to sort to achieve your desired level of reliability; and [c] it improves the probability of getting a representative sample.
 

[a] Discrete information
 

The first reason to stratify by type is to derive separate information about two or more sub-groups of the waste stream for which discrete data is needed to answer your question. A recycling program, for example, will usually be targeted differently for residential and commercial sources. This is because the two groups are distinct and some policies directed at one would have little or no applicability for the other. Also, the entity that collects residential waste (perhaps the municipality itself) will often be a different one from that which collects commercial loads (most often a private hauler). This often results in to two separate programs running in tandem.
 

To generate information for each type of source of waste, it will be necessary to separately sort the two or more sources and compile their data distinctly as well.
 

In the real world, there will usually be a kink in this neat delineation. Apartments with more than 5 units will typically use the same kind of dumpsters as commercial establishments and be picked up in the same truck that collects from stores and factories. Though waste characteristics of apartment dwellers more closely resembles residents of single family homes than that of grocery stores or restaurants, in most cases their waste will be hauled on the same truck as commercial waste.
 

There is no easy resolution to this quandary. Compromises usually entail interviews with haulers to exclude trucks which haul from apartment and commercial dumpsters, but this may inadvertently intrude biases into the results. The only ironclad way to separate the two when separate apartment data is required is to use a waste source strategy (see p. 53).
 

[b] Fewer samples
 

Another reason to stratify is to reduce the number of samples you need to achieve a desired level of reliability. Samples often exhibit a good deal of variability from one to sample to the next compared to their average proportion, and this results in less reliability.
 

However, much of the variation can sometimes be explained because the population actually consists of two or more distinct subgroups, or strata, One group, such as apartments with garbage disposals, might have little food waste, while another, such as restaurants, could have quite a lot. Calculate the average of the two groups, and the percent of food waste in each group's samples will vary substantially from that combined mean.
 

As described in the Stat / _facts primer (see p. 90), stratification permits you to achieve the same confidence interval and level with fewer samples. This is done through pooling techniques that filter out the variation between the strata which is extraneous to the essential uncertainty you need to worry about surrounding your sample results. This sampling strategy will work to reduce the number of samples needed for a desired reliability so long as the variation within your strata is less than the variation between them.
 

Figure 1 illustrates for one particular case how the band of uncertainty, the confidence interval, was narrowed in a typical waste sort by sorting and then pooling residential and commercial strata separately. As you can see, had the analyst not stratified in that particular case, the confidence interval of the combined strata would have been reported as nearly 30%, instead of less than 15%.
 

Often, you should also divide your sampling over different seasons (see p. 17). Although this is primarily done to enhance the representativeness of your results, if the different seasons are treated as if they are different strata in the calculations of confidence intervals, the uncertainty band surrounding the annual averages can be substantially narrowed.
 

[c] More accurate samples
 

Stratification can help get more representative samples than simple random selection can achieve in some cases. Random selection, discussed later (see p. 46), builds upon probabilities. This implies that most of the time random selection will pick samples that are representative of discards from the major sources of waste.
 

However, there will be a small number of times by chance that waste sorts, done with randomization, are not representative of all sources. That will be extremely rare with sources that contribute a major fraction of the waste stream, but rise for sources which are minor. By stratifying the sub-groups which are a smaller part of the whole, you can insure that a full quota from those small groups will be randomly picked.
 

[5] Strategies to reduce biases
 

In addition to the statistics you calculate, the likelihood that your study will produce a reasonable estimate of the correct answer depends upon the success you had getting representative and unbiased samples. In this regard, four key points should be kept in mind when doing a waste sort: [a] seasons; [b] a typical week; [c] hourly and daily differences; and [d] random selection.
 

[a] Selection of seasons
 

In northern climates, it gets cold and snows in the winter. Then comes the spring and everything comes alive...only to die or get cut down, and then get thrown into the landfill. Sample for the proportion of organic material in the waste steam during summer and you will often get a bloated reading on the annual fraction of green matter. Other materials, such as aluminum cans, similarly, vary with hot weather.
 

To reduce those biases, especially in regions with vigorous winters, you should do sampling at least two seasons (winter and summer). Ideally, sampling should be done in all four seasons in order to more accurately reflect the annual distribution of waste.
 

When seasonal sampling is done, the extent to which each season contributes a different proportion of waste to the annual waste stream must be determined so that, when you do your analysis, it can be weighted in computing annual statistics.
 

Incidently, those weights for each seasons that you use in computing annual averages come in handy in calculating confidence intervals, too. The so-called "pooling" techniques, discussed for strata (see p. 14), will permit you to narrow the uncertainty bands around the annual compilation of seasons. For, the only difference between strata and seasons is that, with strata, the groupings are done by type of waste generator, and, with seasons by time of waste generation. Pooling of seasons squeezes out the uncertainty ascribable to variations between seasons for such things as yard waste, aluminum cans, etc., just as it does between strata.
 

[b] Typical Week
 

For each season that you sample, you will only be going out to sort the waste stream for one point in time perhaps for a day or, more likely, for a week. It is important that this point in time typify that season involved. That time ought not have abnormal characteristics from such things as holidays, a time when most people move their residence, or from weather.
 

[c] Hours and days
 

Trucks that unload early sometimes have different type loads than ones later. By using a random number generator to select trucks from which to draw samples (see p. 52), this possible source of bias can be avoided. Randomization insures that every truck during the day has an equal chance of being picked.
 

Variations between days also are theoretically a possible source of bias. But, although this can be serious when studying the total amount of waste generated, it is generally considered less so with waste composition as long as you: (1) avoid holidays and extreme weather; and (2) randomly select trucks in the proper proportion between residential and commercial (and any other significant strata's) loads. Otherwise, you'll need to spread your sampling over several days, and, best yet, over an entire week.
 

[d] Random selection
 

A study design should contemplate that, whenever a choice must be made along the way to getting a sample, random selection procedures are used (for a discussion of that procedure, see p. 27).
 

Selection should never be based upon judgment calls using visual inspection. For example, when 20,000 pounds of smelly garbage from a truck is spread out on the tipping floor, don't eyeball which sample loads "look" representative. Most people would subconsciously avoid the worst of it, even though the waste stream consists, in part, of bad smelling stuff. Instead use the formal random selection procedures that are described in the next chapter Selecting Samples (see p. 45).
 

[6] Materials to sort for
 

A prime question in a waste composition study is which materials to sort for. The chief criterion for selection is the nature of the questions you are seeking to answer with the study.
 

If the study is intended to develop a recycling strategy, then attention will typically be centered on things like glass, metal and plastic containers, newspapers and high grades of paper. If it is to determine the optimal throughput for a proposed incinerator, then the focus will probably be to sort all of the important combustible materials, grouped by ones with different heat and moisture content. On the other hand, if the purpose is to provide information for mixed waste composting, the organic fraction and its components become important, etc.
 

Competing against these considerations are logistical concerns when a list is expanded. True, the bulk of the expense will be incurred setting up and evaluating the sorting process. The cost of adding materials to the list usually is minor in relation to total expenditures. However, errors in measurement are another matter. At some point it becomes cumbersome for the sorters in the field to keep track of all of the materials being separated, and mistakes increase soar. After 30 materials, you should be cautious about further additions unless you have budgeted for careful hiring of qualified sorters, supervising and sorting each sample.
 

Table 2 is an illustrative list of materials for a waste sort

[7] Time periods to sample
 

Sometimes the questions being asked will relate to how the waste composition changes over time (perhaps in order to project the future) or in response to an event.
 

[example: How effective will a new recycling ordinance be at diverting material from the landfill.]
 

[another example: Will the changing ratio of metals to plastic over time significantly affect the heat content of the incoming trash to an incinerator.]
 

Taking sets of samples at different points in time, either over the course of each of several years (called a time series) or before-and-after an event can help provide the data to answer these questions.
 

A before and after research design sorts before the event in question and then another afterwards. Looking ahead to the analysis chapter, a particular type of statistical test, called the significance of the difference between two means test, can be used to interpret the before and after data to determine whether apparent differences are really significant.
 

[8] Degree of reliability needed
 

The big ticket in waste composition studies is how many samples that are costly to sort are needed. That, in turn, will depend upon how much uncertainty you are willing to accept around the averages you calculate from your samples Defining that is an important part of the study design. To figure that out, there are two questions to ask

"How much will it cost if a properly sized system today becomes missized in the future because the composition of the waste stream changes?"
 

Surveys of waste system operators have generally concluded that estimates of waste quantities need to be very close to the correct answer. However, that need has been found less important in waste composition studies for at least two reasons.
 

First, the uses to which waste quantity data is put can be critical to the success of the projects it is used for. Bond repayments in the solid waste industry come from tipping fees at capital intensive waste facilities that have been sized based upon this type of study. In turn, those tipping fees are also calculated from the projections of waste quantities. A wrong guess here could send the whole program into bankruptcy if revenues generated from those tipping fees are inadequate to pay off creditors.
 

With waste composition, on the other hand, misestimation of relative proportions is often less directly related to the project's financial survival. To a significant extent, the bottom line for a materials recovery facility or incinerator will be more closely tied to the total incoming flows, for which a tipping fee is levied, rather than from sales of recyclables or energy. Similarly, were you to incorrectly estimate the proportion of one material, another material's fraction will be overestimated because the total in percentages remains 100%. Many times these offsets in the relative relationship of different materials will come close to netting out to zero in the sales value of the entire stream. This is one reason why the high cost of reliability is usually more difficult to justify for waste sorts.
 

Second, in waste sorts the costs of a wrong guess are spread over the period that a facility is used. Over that time, the relative relationship of many of the materials in the waste stream are likely change from that which you carefully sorted today. This can be due to changes in product container designs, buying habits, and so on. For example, plastic substitutes for glass, and paperboard for some plastic. "Light weighting", or making thinner walled containers, reduces the fraction of all one-way consumer containers. A meticulous breakdown of today's waste stream cannot guard against these fluctuations.
 

That is the reason why the precision of a single snapshot in time is not of overwhelming importance. Overall, a larger degree of uncertainty is acceptable in garbage than in rigorous scientific studies where, for example, maximum safe dosages of drugs are being determined in studies which describe relationships of dose to health that are constant over time.
 

The following table typifies how a desire for greater reliability can translate into drastically more samples to sort. Compare, for example, the number of samples needed in an illustrative case for a wide uncertainty band of ±20 and a low confidence level of 90% here 37 samples to the 844 samples needed for a narrower ±5 ban at the 95% level.
 

[cross-reference: For details on how to estimate the number of samples needed to achieve a desired level of reliability, see p. 31.]
 

In light of these considerations, a reasonable decision would be to use a 90% confidence level and, if funding permits, a ±10% confidence interval. If funding is tight, a ±20% interval could be substituted.
 

All of this has focused on how to randomly select enough samples to get a precise answer. There are other issues interrelated with these. Among them is the physical act in the field of accurately sorting and weighing the sampled materials. For a discussion of good field techniques, see the sorting samples chapter at p. 55. (For a technical discussion of the difference between the statistical concepts of precision and accuracy that underlie the more common term reliability used in the text, see the Stat / _facts Primer at p. 86.)
 

[note: A tear out Study Design Planning Worksheet is included in the appendix. See p. 121.]