Stratified Sampling by Work Shifts
La ingeniería de métodos moderna se enfrenta a una paradoja: mientras la Industria 4.0 promete datos masivos mediante sensorización total, la realidad de la…
Modern methods engineering faces a paradox: while Industry 4.0 promises massive data through total sensorization, plant reality imposes constraints of cost, PLC compatibility, and privacy regulations (GDPR). This is where the Law of Large Numbers and the original technique of L.H.C. Tippett (1934) resurface with critical—but digitized—relevance.
Work Sampling is the standard tool for measuring productivity. However, a common methodological error threatens the validity of these studies in continuous-production systems: the assumption of population homogeneity. Treating the Morning Shift and the Night Shift statistically the same under Simple Random Sampling (SRS) generates a sampling error that blurs operational reality.
WorkSamp proposes an evolution toward Stratified Random Sampling, a mathematical approach that reduces global variance and delivers a productivity diagnosis compatible with the LOPD (Spanish data-protection law) and the upcoming labor regulations of Spain 2025.
Foundations of Stratified Random Sampling in Continuous-Production Environments
The Fallacy of Homogeneity in 24/7 Systems
In industrial statistics, we define a population as homogeneous when the probability of an event's occurrence (e.g., a breakdown or a wait) remains constant throughout the sample. In a plant operating 24 hours, this is false.
The Morning Shift ($S_1$) and the Night Shift ($S_3$) are, for the purposes of statistical inference, divergent populations:
- Shift $S_1$ (Morning): Characterized by high administrative load, management interruptions, and strong engineering support.
- Shift $S_3$ (Night): Defines an ecosystem of less supervision, biological circadian fatigue in operators, and, critically, less auxiliary logistical support.
Applying simple sampling aggregates these variances, hiding each schedule's specific inefficiency "peaks" under a misleading average.
Precision Mathematics: Variance Reduction ($V$)
The technical superiority of stratified sampling lies in its ability to isolate the within-stratum variance. By calculating the specific sample size ($n_h$) for each shift (stratum $h$), we minimize the variance of the global estimator ($\bar{p}_{st}$).
The formula that governs WorkSamp's precision in this context is:
$ V(\bar{p}_{st}) = \sum W_h^2 \frac{p_h (1-p_h)}{n_h} $
Where $W_h$ represents the stratum's weighted share (shift hours relative to the total).
The empirical impact is direct: Internal WorkSamp data show that optimal stratification allows reducing the total sample size ($N$) by 12% to 15% while maintaining the same Confidence Level ($Z=1.96$ for 95%), or alternatively keeping $N$ constant and significantly reducing the margin of error. This translates into faster and more cost-efficient diagnoses.
WorkSamp Methodology: Statistical Rigor and MECE Taxonomy
For inference to be valid, data collection must follow strict protocols that eliminate ambiguity.
MECE Taxonomy to Prevent Data Overlap
The activity classification system must be MECE (Mutually Exclusive and Collectively Exhaustive). Any overlap between categories invalidates the inferred OEE calculation. WorkSamp standardizes three macro-categories:
- VA (Value Added): Physical or chemical transformation of the product. It is the only time the customer pays for (Wrench Time).
- NVAN (Necessary Non-Value-Added): Inevitable support activities with current technology (transport, quality inspection, reading drawings).
- NVA (Waste): The focus of improvement. Includes waits, breakdowns, unnecessary movements, and idle time.
Neutralizing the Hawthorne Effect with 'Snap Reading'
The Hawthorne Effect describes how individuals modify their behavior when they know they are being observed. Traditional stopwatch measurement exacerbates this bias.
WorkSamp uses the Snap Reading technique: asynchronous random observations. When stratified by shifts, the algorithm generates routes where the observation probability remains proportional even at critical hours (e.g., 03:00 AM – 04:00 AM). This ensures the data reflects the real process, not "audit behavior," and eliminates the observer's convenience bias.
Empirical Case Analysis: The "Black Box" of the Night Shift
Below we present data extracted from the Technical Report 2025-WS-STRAT on a plastics-injection plant, demonstrating why the simple average is dangerous for decision-making.
Comparison of Wrench Time and Micro-Stops by Stratum
| Metric | Morning Shift ($S_1$) | Afternoon Shift ($S_2$) | Night Shift ($S_3$) | Average (if SRS) |
|---|---|---|---|---|
| Wrench Time (VA) | 42% | 38% | 29% | 36.3% |
| Waiting for Material | 15% | 12% | 22% | 16.3% |
| Micro-stops | 5% | 7% | 18% | 10% |
Technical Interpretation:
A Simple Random Sampling (SRS) would have reported an average Wrench Time of 36% and material waits of 16%. The Operations Director might conclude the plant has an "acceptable" performance.
Stratification, however, reveals a critical productivity drop on the Night Shift (29%) driven by a specific logistics failure: 22% of the time is lost waiting for material (probably due to a lack of forklift drivers on that shift).
Impact on Energy Cost and Sensorless OEE
This night-time inefficiency is financially toxic. Although the night-time electricity tariff is lower, the opportunity cost of having injection machinery at low OEE is greater.
WorkSamp allows calculating a Sensorless OEE (Inferred OEE) without connecting a single cable to the PLCs:
- Availability: Inferred from the % of observations in "Breakdown" or "Setup."
- Performance: Cross-references the % of VA with the theoretical output.
- Quality: Measured by the frequency of "Rework" or "Scrap" observations.
2025 Regulatory Context: Why Human Sampling Beats AI
On the Spanish industrial horizon, invasive technology runs into insurmountable legal barriers.
Privacy, GDPR, and Union Resistance
Deploying computer-vision cameras to measure cycle times collides head-on with the GDPR and resistance from Works Councils. WorkSamp offers an absolute legal advantage:
- Anonymity: Data is tied to the work center or machine, never to the individual.
- Process vs. Individual: Snap Reading audits the process flow, strictly complying with the LOPD.
Adapting to Working-Hour Reductions
With the trend toward a 37.5-hour workweek, nominal production capacity is reduced. The only way to maintain output without raising labor costs is to increase real work density—that is, raise Wrench Time. Only what is measured with stratified precision can be improved.
WorkSamp Solution: Scientific Process Audit
WorkSamp is not just software; it is a digitized statistical-inference methodology designed for engineers who demand rigor.
Technical Capabilities:
- Automatic N Calculation: Algorithms that determine the sample size needed to ensure an error $<\pm 3\%$.
- Stratified Routes: Generation of weighted random observation schedules by shift.
- Real-Time Analytics: Dashboards that visualize the Gaussian curves of productivity and data dispersion instantly.
Efficiency is not guessed—it is statistically inferred.
Request a Statistical Feasibility Diagnosis for your Plant.
Validate your bottleneck hypotheses and uncover hidden productivity in your shifts with no hardware investment.
[Contact WorkSamp Engineering]
Technical Frequently Asked Questions (FAQ)
1. What is the minimum sample size for the study to be valid per shift?
To guarantee a 95% Confidence Level ($Z=1.96$) and a margin of error of $\pm 5\%$, statistical theory dictates a minimum of 384 valid observations per stratum (shift). To reduce the error to $\pm 3\%$, approximately 1,067 observations per stratum are required. WorkSamp calculates this automatically based on the total population and expected standard deviation.
2. How is Stratified Sampling different from Simple Random Sampling on the plant floor?
Simple Sampling fires random observations over total operating hours without distinction. Stratified Sampling divides the day into sub-populations (shifts) and assigns specific observation quotas to each one. This prevents—purely by chance—too many samples being taken in the morning and too few at night, which would skew the overall results.
3. Is it legal to conduct Work Sampling under the GDPR in Spain?
Yes—and it is preferable to video surveillance. The WorkSamp method is based on observing the activity of the machine or work station, not on biometric tracking of the operator. By not recording identifiable personal data, it complies with the LOPD and reduces friction with workers' representatives.