Work Sampling in Waste and Recycling Plants

If you manage a municipal or industrial waste treatment plant, you know that a bottleneck can be organisational as well as technical. Conveyors run, optical sorters classify material, balers compact it, and manual sorting cabins are staffed with people in protective clothing and gloves. Yet the amount correctly separated can still fall short of expectations. The source example mentions a 12–30% range, but without a defined waste stream, denominator or supporting dataset it should not be treated as a sector benchmark. Rigorous observation can help establish where the shift goes; it cannot, by itself, explain every difference in recovery yield.

This article explains how to apply work sampling—the statistical observation technique associated with Tippett's work in the British textile industry in the 1930s—to a waste treatment plant. It is a practical guide to what to measure, how to size the sample, which observation categories fit sorting, maintenance and recovery operations, and which operational decisions a well-designed study can support.

1. Why a stopwatch alone does not describe a waste plant

Stopwatch time study works well for repetitive, predictable operations that can be divided into meaningful elements. Many activities in a waste plant are less regular. A manual sorter encounters different items, optical sorting cycles are machine-driven, and maintenance includes waiting, breakdowns and decisions affected by the incoming material. A stopwatch can still be useful for selected operations, but an element-level study alone may miss how an entire shift is used.

Work sampling, by contrast, estimates how frequently a system—an operator, machine or team—is in each defined state, rather than timing every task. The central question is not simply “How long does this take?” but “What proportion of the observation period is spent sorting, waiting, maintaining equipment or performing other activities?” That is often the more useful capacity question for a waste treatment plant.

In its standard form, an observer—or a suitably validated automatic capture system—records the state of each sampled person or asset at randomly selected times. From a set of observations, the study estimates each state's share of time and reports uncertainty. The confidence level and target precision are specified at the design stage and the achieved precision is checked afterwards. Dependence between observations must be considered rather than assuming that every recorded state is independent.

2. What changes in the waste sector

Work sampling has been used in manufacturing, maintenance, offices, hospitals and logistics. Waste treatment has three characteristics that require the category system and sampling design to reflect the plant itself.

Incoming loads can vary substantially. Municipal waste composition can change during a day, week or season. Holiday collections, summer packaging and seasonal garden waste are possible influences, not universal patterns. This makes it important to stratify by shift, day of the week or incoming material type where those distinctions answer the study question. The sampling period must cover the conditions to which the conclusions will apply.

Safety determines how observations can be made. Required personal protective equipment is a safety condition, not an assumption that shortages never occur. A stoppage caused by a missing safeguard must be recorded and addressed safely. Other useful states include an upstream conveyor stoppage or a full receiving container. Categories must describe the actual process rather than importing a generic catalogue from another industry.

Contractors can be part of the operating model. Where sorting, maintenance or cleaning is outsourced, the analysis may need to distinguish plant-controlled constraints from contractor-controlled arrangements. For example, material shortages, scheduling, staffing coverage and training can require different follow-up. Observation alone does not assign contractual fault: records, responsibilities and the circumstances of each event must also be reviewed. For a large operational audit, this distinction can be important.

3. Observation categories for a sorting and separation plant

The following eight-category framework is an illustrative starting point for plants processing roughly 50,000–200,000 tonnes per year. It is not a validated standard for every plant. Adapt it after preliminary observation, define mutually exclusive states and resolve overlaps before data collection.

Value-adding activity (VA): manual separation, operating optical sorting equipment, removing contaminants, preparing recovered-material orders, feeding the baler or loading a hopper with a loader. These are the direct operating activities defined as VA for this study; the label does not judge the worth of a person or of necessary supporting work.

Available at the workstation but waiting (VA-ESPERA): an operator is correctly positioned but has no material to process because of an upstream interruption, blocked flow or a full destination container. Despite its legacy code, this category records waiting rather than active value creation. A moving conveyor does not imply that usable material is available.

Necessary non-value-adding activity (NVA): walking to a workstation, adjustments, cabin cleaning, operational discussions and incident recording. Clearing a blockage must follow the site's isolation and safety procedures. Necessary work should not be equated with avoidable waste.

Breakdown-related downtime (IMP-AV): a mechanical, electrical, hydraulic or software failure stops the relevant process. Instrumented plants may capture it automatically. Otherwise, the observer should confirm the event using the agreed definition and available records; seeing maintenance personnel is not enough to establish a breakdown.

Planned-maintenance downtime (IMP-MP): scheduled preventive or predictive maintenance and inspections. Distinguish it from a breakdown. It is planned work whose timing and impact can be assessed without assuming it should be eliminated.

Material-related downtime (IMP-FM): lack of feedstock, unsuitable material for the scheduled process, or an unloading-area bottleneck that prevents the truck from discharging. Specify whether the observation refers to the plant, a line or a person.

Staff availability or assignment (IMP-P): an uncovered absence, an operator awaiting assignment, or a staffing mismatch with the flow. Meals, authorised breaks, shift briefings, sickness absence and access restrictions are different circumstances and should be separated where relevant. They must not automatically be classified as avoidable personal inefficiency.

Other states (IMP-O): anything not captured elsewhere. A share above an illustrative 3–5% review threshold can prompt a check of definitions or training, but the threshold is a project convention, not a statistical law. Some “other” activity may be legitimate and informative.

4. Sizing a sample in a plant with 80 operators

For a plant with 80 people across sorting, maintenance and recovery, the observation requirement starts with the proportion to be estimated. A common normal-approximation formula for an independent simple random sample is:

n = (z² · p · (1 − p)) / e²

At 95% confidence, z = 1.96. With p = 0.5, the conservative maximum-variance assumption, an absolute margin of error of 5 percentage points gives 385 observations when rounded up. A margin of 3 percentage points gives 1,068 when rounded up; the often quoted 1,067 is the unrounded 1,067.11 result rounded to the nearest integer, not the conservative upward rounding.

The requirement applies to the population or subgroup whose proportion you want to estimate—not automatically to every employee. If the aim is a plant-level estimate, design observations across the relevant people and times. If separate, equally precise estimates are required for every person or asset, each estimate needs an appropriate sample. Stratification, unequal weights, repeated observations and clustering may change precision and sample allocation.

The source scenario uses 1,067 observations for each of 80 people, or approximately 85,000 person-state records (1,067 × 80 = 85,360). That is an illustrative individual-level design, not a universal minimum. It does not imply more than 1,400 observer-days: field effort depends on route duration, the number of people recorded on each round, safe access and temporal coverage. Several simultaneous observers or validated automatic capture may be useful, but staffing cannot be calculated from the record count alone.

A study with 200 independent observations at p = 0.5 has an approximate 95% margin of error of 6.9 percentage points. That may or may not be sufficient for its decision. Audit suitability depends on the question, design, bias controls and uncertainty, not on a blanket rule that any sample below a certain size is unusable.

Operational option: photocells, belt scales, SCADA and container readers can supply some equipment states. Optical sorters may report operating status, balers their cycles and stoppages, and conveyors their speed. Human activity still needs an appropriate recording method, whether human observation or a properly validated system. Computer vision requires safety, privacy, employment-law and data-governance assessment as well as sufficient classification accuracy.

5. Decisions a well-designed study can support

Work sampling is an input to operational decisions, not an end in itself. The following numerical examples are hypothetical, not measured WorkSamp customer results.

Reviewing an outsourced sorting contract. If 22% of sampled time is coded as staffing-related unavailability, management can investigate coverage arrangements and discuss them with the contractor. Before considering penalties or price changes, separate authorised breaks and other obligations from uncovered absence, establish causes and review contractual responsibilities. The percentage alone does not establish fault.

Assessing manual sorting capacity. A 35% share coded as waiting can identify a flow or availability problem. It does not, by itself, show that a conveyor is too fast. Excessive speed and an inability to pick available material are a different condition and need their own observations, throughput and recovery-quality measures. Depending on confirmed causes, options may include speed changes, additional manual sorting positions or improved upstream optical separation.

Supporting an additional optical sorter. If a comparison of parallel lines shows 18% more VA on the line with extra optical sorting, it can inform an investment proposal. Check whether “18%” means a relative increase or percentage-point difference, and whether feedstock, staffing and scheduling are comparable. An observational difference is not proof that the extra sorter caused the gain.

Improving maintenance scheduling. If planned maintenance repeatedly occurs on Monday mornings after a weekend shutdown and coincides with feedstock congestion or material-related waiting, compare the timestamps and process records. The combined evidence may identify a scheduling opportunity; two codes appearing in the dataset are not sufficient to establish the cause.

Supporting recycling improvement without claiming regulatory certification. Directive (EU) 2018/851 amended Directive 2008/98/EC and introduced municipal-waste preparation-for-reuse and recycling targets of 55% by weight for 2025, 60% for 2030 and 65% for 2035, subject to the legislation's conditions. These are not required VA percentages for an individual plant. Work sampling can support process improvement, but does not demonstrate compliance with recycling targets or replace the prescribed waste-flow measurements and reporting.

6. Simulated practical example

The following case is constructed for teaching purposes using illustrative sector assumptions. It is not an actual WorkSamp implementation or a specific customer case. Its costs and expected improvements are scenario inputs, not validated market benchmarks.

A municipal solid waste (MSW) sorting plant in a medium-sized city processes 120,000 tonnes per year and has 84 operational staff across three shifts. The source scenario assumes 1,067 observations per person at a nominal 95% confidence level with p = 0.5 and an approximately 3-percentage-point margin, spread over 12 full study days: four per shift, alternating weekdays and weekends. For a new study, round the simple formula up to 1,068 and assess whether the proposed rounds can provide the required coverage and sufficiently independent information.

Aggregated results, using weights appropriate to the numbers of operators and the sampling design, are:

Category	Percentage of time
Value-adding activity (VA)	41%
Available but waiting (VA-ESPERA)	17%
Necessary non-value-adding activity (NVA)	14%
Breakdown-related downtime (IMP-AV)	8%
Planned-maintenance downtime (IMP-MP)	5%
Material-related downtime (IMP-FM)	9%
Staff availability or assignment (IMP-P)	4%
Other states (IMP-O)	2%

Operational interpretation: 59% of time is outside the narrowly defined VA category, not necessarily time that can or should be eliminated. Planned maintenance, waiting and necessary NVA account for 36 percentage points (5 + 17 + 14). Breakdowns, material shortages and staffing-related states account for 21 points (8 + 9 + 4), while the remaining 2 points are “other”. Whether any of these states is avoidable must be investigated.

For 84 people at an assumed average loaded labour cost of €32,000 per person per year, the base is €2,688,000. The 21 points represent approximately €565,000 per year of labour-cost-equivalent time. This is not automatically cash that can be saved, nor does the category split establish that the whole amount is unproductive or recoverable.

Three hypothetical actions illustrate how to evaluate the data:

1. Reduce IMP-FM from 9% to 5% through better coordination with municipal collections. Four percentage points of the €2.688 million labour base are approximately €107,000 per year of time capacity. The cash effect depends on whether that capacity changes overtime, subcontracting or output and whether the identified causes are removed.
2. Reduce IMP-AV from 8% to 5% with a vibration- and thermography-based predictive maintenance programme. The scenario assumes a programme cost of approximately €45,000 per year and a gross annual benefit of approximately €80,000. The difference is €35,000 before other costs. Dividing €45,000 by €80,000 gives roughly 6.75 months, but this is not a valid investment payback unless the €45,000 is an upfront investment rather than a recurring annual cost and the benefit assumptions are justified.
3. Address the 4% IMP-P category with rapid absence coverage. The scenario assumes a system cost of approximately €25,000 per year and a gross benefit of approximately €35,000 per year, giving €10,000 before other costs. The quoted payback of under nine months similarly requires an upfront-cost assumption; an annual operating expense must be included in the net benefit calculation. Coverage also creates a cost and does not necessarily remove a full four points of unavailable time.

The three gross scenario benefits total €222,000 per year, around 8.2% of the approximately €2.7 million labour base. After the two stated recurring programme costs, the figure is about €152,000 before any additional implementation costs and before testing overlap or cash realisation. These are potential benefits, not achieved savings. A claim that the study pays for itself in fewer than five months cannot be checked without a specified study cost and benefit profile.

7. Practical delivery: from study design to the report

A waste-plant work-sampling study can be organised into five phases. The durations below are planning examples, not guaranteed delivery times.

Design and planning: one to two weeks. Write down the specific questions. “What is the plant's overall activity split?” is different from “How do contracted and in-house sorting activities differ under comparable operating conditions?” The question determines categories, sampling units, sample size and period. Decide whether observation will be continuous over a period, with random observation times, or a focused two- to three-week campaign with several observers.

Preliminary observation and category validation: about one week. Observers spend time across shifts, identify real operating states and test the proposed category definitions. A high “other” share—for example, 15%—is a reason to investigate definitions and training, not automatic proof that the entire study is invalid. Revise and pilot the categories before the main campaign.

Data collection: two to six weeks, depending on resources. Observers use recording sheets or tablets and record each sampled person or asset at random times. A route is defined in advance, checked for safe access and followed consistently, while potential route-order bias and the delay between the planned time and actual observation are addressed. A random interval such as 8 ± 2 minutes is an operational illustration; the actual schedule must not systematically align with process cycles or be easily anticipated.

Statistical analysis: about one week. Estimate the share in each category and report appropriate confidence intervals. Analyse shifts, areas and material types where the sample supports those estimates. Investigate differences and translate time into labour-cost equivalents using area-specific costs. Keep sampled activity, root-cause evidence and avoidable cost separate.

Report and action plan: about one week. Present results to plant management and, where relevant and authorised, to the municipal-service contractor. Prioritise actions with estimated costs, benefits, timescales and owners. A report should show how the data support each recommendation and what remains uncertain.

8. Study-quality checklist

Before accepting the study, check these twelve points:

1. The exact question is written unambiguously in the report.
2. Categories have been piloted and approved by the plant representative, with operational definitions.
3. The sample-size calculation applies to the population or subgroup reported, with the chosen confidence level, precision and design assumptions.
4. Observation times are random across relevant operating periods and are not systematically anticipated.
5. The observation route is preplanned, safe and checked for route-order and timing bias.
6. Shift, day and material-type stratification reflects the study question.
7. Confidence intervals are reported for the principal category estimates.
8. Labour-cost equivalents are stated in euros per year, with assumptions; they are not labelled automatically as cash savings.
9. Each action has an owner, deadline and follow-up KPI.
10. A repeat study is planned when appropriate—for example, after six months—to check whether improvements persist.
11. In-house and contracted activities are distinguished where relevant, under comparable conditions and with justified weights.
12. The report assesses the operating system and structural causes, not unsupported judgments about individual workers. Privacy, employment rights and required breaks are respected.

9. When work sampling may not be the right next step

When the plant has fewer than 30 people. The source uses 30 as an illustrative screening figure, not a statistical minimum. Study economics depend on the question and risk, not headcount alone. A smaller plant may start with structured process observation; a statistically designed study can still be appropriate if the decision warrants it.

When management will not act on the results. Before commissioning a study, agree who will review it, prioritise actions and assign owners. Otherwise even sound evidence may deliver little operational benefit.

When the plant is being redesigned or closed. If the line configuration will change shortly or a concession is ending, establish whether a baseline, transition study or later campaign would actually inform a decision. Data are not automatically useless, but their intended use must be explicit.

10. Conclusion

Work sampling can be particularly useful in waste treatment because a major question is how people, assets and flows use the available time—not just how long one repeatable task takes. Its statistical foundations are established; sector application requires representative sampling and categories that match the plant, rather than a claim that the same design fits every site.

A plant-specific budget must include observer time, training, software if used and implementation costs. Estimate benefits from verified causes and realizable changes; neither headcount nor an activity percentage establishes a market price, savings range or payback.

For recycling improvement, observation can complement process evidence. It does not certify compliance with the Directive's 55%, 60% and 65% municipal-waste targets or establish a required VA share for reporting to Spain's Ministry for the Ecological Transition (MITECO).

The next step is to define the question the plant needs to answer and select an appropriately experienced team. Relevant process knowledge matters alongside statistical competence and safe access arrangements.

If you want a tool for study setup, observation recording, sample-size calculations and reports, start with WorkSamp. Confirm and adapt the category framework to the site rather than assuming a generic taxonomy is ready for every plant. For repetitive operations outside the manual sorting task, such as a baler cycle, Cronometras provides a digital stopwatch with configurable ILO and TAL allowances. If the challenge is attendance recording across several contractors, Induly supports digital working-time records. The ASETEMYT directory can be consulted for industrial organisation professionals and companies; check relevant experience and credentials before selection. These services form part of the ProdCont industrial-productivity ecosystem for measuring, planning and making data-informed decisions.

---

Sources and technical scope:

• INE, Waste Accounts. Use the release's actual reference year: the official page consulted reports 112.7 million tonnes generated in 2023, not a verified “110.1 million tonnes in 2024” figure. INE waste accounts.
• Directive (EU) 2018/851 of the European Parliament and of the Council of 30 May 2018 amending Directive 2008/98/EC on waste. The targets concern preparation for reuse and recycling of municipal waste by weight, not sampled labour-time shares. Official text in the BOE.
• NIST/SEMATECH e-Handbook of Statistical Methods, “Selecting Sample Sizes”: objectives, precision, variability and sampling design determine sample requirements. NIST.
• Tippett, L. H. C. (1935), “Statistical Methods in Textile Research. Part 3—A Snap-Reading Method of Making Time-Studies of Machines and Operatives in Factory Surveys”, Journal of the Textile Institute Transactions, volume 26. Historical reference supplied by the source article; no claim of a separate verified 1927 origin is made here.

Technical reference.