The 'Other' Category: Risks and Management

Why the "Other" Category Destroys the Validity of Your Work Sampling Study

In a Work Sampling study, every observation is data that must fit into a logical and robust classification system. The "Other" category is not a neutral option; it is an admission of methodological defeat. It represents the point where our measurement system fails to capture operational reality, creating a statistical "black box" that corrupts from the root any inference about productivity.

The solution is not to settle, but to apply the same statistical rigor that we use to calculate sample size or confidence level to design an impeccable taxonomy. This article is your technical guide to achieve this.

The MECE Principle: The Backbone of a Valid Taxonomy

A useful taxonomy must be MECE: Mutually Exclusive and Collectively Exhaustive. This means that each observed activity must fit into one and only one category, and that among all categories they must cover 100% of the possibilities.

The "Other" category directly breaks the exhaustiveness principle. It tells us nothing. It is a catch-all that can contain anything from a micro-stoppage due to lack of material to an unproductive conversation or an urgent corrective maintenance task. Mixing these realities prevents any root cause analysis.

When is an "Other" legitimate? Only during the pilot phases of a study, and it should never exceed 5% of total observations. If your definitive study has a persistent "Other", your taxonomy needs urgent revision.

Sector Benchmark: Where Do We Stand?

Data from industrial plants in Spain reveal that this is an extended and quantifiable problem. According to multisector studies, typical percentages of the residual category are alarmingly high:

Automotive: 11.2% ± 4.3
Pharmaceutical: 8.7% ± 2.1
Agri-food: 18.4% ± 6.7
Metal-mechanical: 14.6% ± 5.9

These numbers are not just a statistical curiosity. Each percentage point above the 5% threshold introduces noise, distorts reality, and undermines the study's credibility before management and operators.

Quantified Risks of Maintaining an "Other" Category Above 10%

Ignoring an inflated "Other" is not an option. It carries tangible risks that affect technical validity, operational precision, and regulatory compliance.

2.1. Statistical Invalidation of the Study

The Tippett technique, foundation of Work Sampling, assumes that the data fit a binomial distribution, which can be approximated by a Gauss Curve for large samples. The "Other" category violates the basic premises of this model.

Rupture of the Tippett exhaustiveness principle: The technique requires that the sum of probabilities of all categories be 1. A significant "Other" means that a portion of the distribution is unknown.
Type II error in inferences: You increase the risk of concluding that there is no difference in productivity between shifts or methods, when in reality the difference is masked within "Other".
Artificial inflation of the confidence level (Z): By not capturing all the real variability of the system, your confidence interval calculation will be too optimistic. You will believe you have more precision than you actually have.

2.2. Distortion of the Real Wrench Time

Wrench Time—the time the operator or technician dedicates to working directly on the product—is a key efficiency metric. Up to 40% of activities classified as "Other" are, in reality, value activities that could be reclassified with a more granular taxonomy.

Hidden value activities: It is estimated that between 5 and 12 percentage points of real Wrench Time may be "hijacked" in the other category.
Impact on OEE without sensors: The calculation of Overall Equipment Effectiveness (OEE) through sampling depends on accurately classifying Availability, Performance, and Quality times. Unclassified micro-stops (adjustments, quick cleaning, tool searches) distort the calculation of Availability and Performance.

2.3. Regulatory Non-Compliance

In the current Spanish regulatory landscape, lack of methodological rigor has consequences.

Law 17/2022 on Market Unity Guarantee (Art. 24): Requires that measurement methodologies be "non-discriminatory and technically justified". A high residual category can be interpreted as a lack of technical justification.
ISO 9001:2015 + A1:2024 (points 7.1.3 and 9.1.1): Requires that measurement methods provide "valid, repeatable, and reproducible" results. An "Other" >10% directly compromises these requirements.
UNE 66177:2024 Productivity Management: Explicitly establishes that categories must cover more than 95% of observed time to be considered valid.

Calculation of the Impact on Your Target Margin of Error

This is where statistical theory meets practical reality. The base formula for determining sample size (N) in a Work Sampling study is:

N = (Z² * p * (1-p)) / e²

Where:

Z = Z-value for the desired confidence level (e.g., 1.96 for 95%)
p = Estimated proportion of the activity of interest
e = Acceptable margin of error

The problem arises when p is uncertain due to the "Other" category. If your goal is to measure productive time with a margin of error (e) of ±3%, but you have 15% of observations in "Other", your initial calculation of N is already tainted.

Practical example:

With an "Other" of 15%, the real margin of error can inflate to ±4.7%.
This means that a reported Wrench Time of 65% could actually be as low as 60.3% or as high as 69.7%. A 9.4% variability that makes the data useless for decision-making.

The correlation between the percentage of "Other" and the distortion of OEE calculated by sampling is significant (r = 0.73, p<0.01). The more "Other", the less reliable your global efficiency indicator.

Technical Protocol to Eliminate the "Other" Category and Recover Validity

The solution is a systematic approach combining taxonomic redesign, statistical adjustments, and, where possible, non-invasive supporting technology.

Phase 1: Taxonomy Redesign with Factor Analysis

Do not design the categories at a desk. Analyze your historical data.

Conduct an exploratory factor analysis on a large set of preliminary observations (>500). This statistical analysis will identify natural "clusters" of activities that tend to occur together.
Establish a minimum of 7 primary categories that are clearly MECE. A robust scheme usually includes:
- Operation (Value Added)
- Adjustment/Setup
- Wait (Material, Information, Machine)
- Internal Transport
- Inspection/Control
- Unplanned Maintenance
- Organizational Unproductive Time
Implement digital decision trees on observers' tablets. An app like Cronometras can guide the observer with logical questions ("Is the machine in automatic cycle?" -> Yes -> "Operation") to force correct classification and eliminate the mental shortcut of "Other".

Phase 2: Adaptive Stratified Sampling

If during the first 200 observations the "Other" category exceeds 10%, activate a corrective protocol:

Increase the sample size (N) using the Tippett formula, but introducing the proportion of "Other" (p_other) as an additional uncertainty factor.
Implement directed convenience sampling: Dedicate a percentage of observation rounds (e.g., 20%) to focus specifically on activities that observers doubt how to classify. The goal here is not random sampling, but understanding for reclassification.
Use the Snap Reading technique with non-uniform intervals: Alternate between observations at random moments and scheduled observations to capture cyclical or atypical activities that often end up in "Other".

Phase 3: Non-Invasive Technology Integration

Modern technology can close the classification gap without the need for invasive sensors on machines.

Tablets with contextual validation: Tools that use the device's GPS and gyroscope to suggest categories based on location (adjustment zone vs. production zone) and observer movements.
Anonymous video analytics: Systems that process images from existing cameras to classify large activity categories (person at workstation, person transporting, inactive person) without identifying individuals. This macro-classification can serve as a basis to validate or question manual observations.
Correlation with MES/SCADA systems: For plants with execution systems, it is possible to cross-reference the timestamps of "Other" observations with machine events (stoppages, alarms, state changes) to infer their real cause. Production control platforms like Induly facilitate this integration by centralizing production and timekeeping data.

Advanced Statistical Framework for Persistent Cases

When, despite everything, "Other" persists in studies with N > 1000, it is time for more complex models:

Gaussian Mixture Models: Assume that the "Other" category is not homogeneous, but contains sub-populations (hidden sub-categories) with their own distributions. This model can identify how many sub-categories there are and what proportion they represent.
Bayesian Bootstrapping: A resampling technique that allows estimating credibility intervals (not just confidence) for the proportion of each residual activity, offering a more honest measure of uncertainty.
Statistical Process Control (SPC) Limits: Treat the proportion of "Other" in each batch of 50 observations as a process metric. If it exceeds the upper control limits, an immediate review of the taxonomy and procedure is triggered.

Resources and Tools

To implement these methodologies, it is essential to rely on specialized tools and knowledge communities:

WorkSamp: Specialists in Work Sampling, offering the platform and methodological expertise to execute studies with the statistical rigor described here, eliminating root problems such as the "Other" category.
Cronometras: Essential software for time and motion analysis, with features that facilitate the creation of structured taxonomies and standardized data collection.
Induly: Production Control and Industrial Timekeeping platform that allows integrating sampling data with production events in real time.
ASETEMYT Directory: Find providers, consultants, and specialized tools in industrial timekeeping and methods engineering.
ASETEMYT Blog: Technical articles and case studies on productivity and work measurement.
Have a project or tool? Add it to the directory and help the community grow.

Methods engineering and time study are not disciplines of the past. They are the quantitative foundation of modern productivity, evolving with tools that give them more precision and relevance than ever. Eliminating the "Other" category is not a bureaucratic exercise; it is the foundational act to build a management culture based on real data, not assumptions.