WORK SAMPLING
(Solving problems begins with knowing that there is a problem)
 

A need for current information

Cost information on a construction project is usually tabulated on a weekly or monthly basis.  Costs related to materials are usually reported monthly while labor costs are reported weekly.  This points out the important role that contractors place on labor costs. Labor is the one area over which the contractor's control can make a significant difference.  Costs of materials and subcontract prices are often determined exactly or with considerable accuracy prior to the beginning of construction.  At any rate, substantial profits (or losses) can be realized depending on the management of the labor resources on the job.

In spite of the need for additional emphasis on controlling labor, it is clear that a weekly report is still not effective in truly controlling those costs.  A more timely acquisition of information is needed.  With weekly reports it is still often too late to affect change in labor costs.  This is due to the nature of construction which is dynamic and continuously changing.

Daily cost reports are not feasible on most projects.  Even if they were, they may not give information that highlights problem areas or areas where productivity is lagging. For example, if the estimate is in "error", the costs for the project are compared against unreliable data.  By relying on the estimate, one does not focus on maximizing profits. These cost reports, as mentioned, are usually generated too late to be effective.  In addition, it must be recognized that the job costs incorporate a variety of factors (quality of management, variations in climatic conditions, site conditions, equipment performance and availability, and labor output).  Thus, another means of control, preferably one that can isolate the effectiveness of labor, is desired.  One such technique is known as "work sampling".  Work sampling is effective and timely (corrective action might be taken before the work task is history).  It is "management by exception" in that it focuses on "non-standard" behavior on the project.

Job cost reports have other shortcomings.  The job cost reports are not a measure of the efficiency with which resources are managed.  They are simply a measure of how the job costs "stack up" against the estimate.  Ideally it would be good to have a more accurate measure of the utilization of the resources (labor, equipment, materials, money, time). Of these resources, labor is of the greatest importance as the labor component is often the most expensive, and is most responsive to managerial control.
 
 

 The need for a statistical basis

If management were to have full knowledge of the performance of the labor resource on the project, continuous observation would be required.  This would require several "observers" and would undoubtedly be prohibitively expensive.  This quickly points out the need for sampling.  Sampling is used to establish a measure of productivity that generates data that contains enough detail with which management can make enlightened decisions.

While continuous observations are not practical, too few observations may not be meaningful.  Suppose that an "observer" was asked to walk onto a construction site and note (at the instant of seeing a worker) if the first worker seen was working or not working.  Clearly, the information would not be meaningful.  If the worker who was observed was working, then it would be inferred that everyone on the job was working. This information would be so generic as to provide no help to management.  Even if the observer assessed the status of two workers the information would be too weak to justify any type of action.  However, it would be better than the information obtained from only one worker.  Clearly, more observations should be made.  The required minimum number of such observations can be determined by using statistics.  It must be recognized that exact information cannot be obtained through sampling, but it will be "close enough" for management to be more effective.

Observations made for work sampling result in two possible outcomes (working or not working).  The observer's assessment is a binomial distribution.  This is therefore known as a Bernoulli variable.  The probability of observing a worker in a state of working is known as "p" and the probability of observing a worker in a state of not working is known as "q".  Thus, p + q = 1.  This is true of mutually exclusive events, i.e., a worker cannot be considered working as well as not working.  The key number being sought is "how many or what percent of the workers on the project are working?"
 

Field Ratings

Work sampling can be done in a variety of ways.  The technique used should be independent of the number of individuals to be observed on the job.  Some general rules on work sampling will be given.

The "counting" can be done in several ways.  One way is to make notes on a clip-board note pad.  Another way is to use two mechanical counters (one to count those "working" and the other to count those "not working".

It is preferred that all personnel be sampled.  If data are collected in sufficient detail, a differentiation can be made between members of different crafts, those working in specific areas, and even those working for different craft supervisors.

The "observer" should do nothing else while in the process of making the observations. Assessments of "working" or "not working" should be made at the very instant that the worker is first seen.  Do not make guesses as to the activity being performed.  It is important to minimize bias.

The observer should be carefully trained in the area of performing work sampling studies.  The observer should know how and why the data are being collected.

Randomize the timing of sampling periods and the routes followed when making observation tours.  Observations are not to be made during breaks or shortly before or after them.  Neither should they be made during the first or last 30 minutes of the work day.  Sampling routes can be randomized by simply having the start points determined by a random number generator of some type.

What is "working"? It is generally considered as performing a task that results in the physical completion of a project.  It includes carrying materials, holding materials, supporting material, participating in active work, discussing work plans, etc.

What is "not working"? It includes waiting, talking (not job related), walking empty-handed, riding, not involved in a useful task (featherbedding may a part of the union work rules).  It includes those activities or inactivities that do not contribute to the final completion of the project.

What is "percent working"?  It is simply the number of observations of "working" divided by the total number of observations.  Some writers on this topic advocate the addition of 10% to the derived number in order to allow for personal time and time spent with a supervisor as in getting instructions.  The 10% allowance appears unwarranted, however, as the same 10% would be added each time.  Thus, in order to simplify the calculations it is advisable to exclude the 10%.

It is important that the results of work sampling studies should not be used for disciplinary purposes or for dismissals.  The effectiveness of the tool will be lost if it is used in this way.  The ratings are equally a measure of managerial effectiveness as they are of the effectiveness of work crews.  They are not a measure of the effectiveness of individual workers.
 
 
 
 
 
 
 
 

 Definitions related to work sampling:

 w = Number of workers assessed as "working" (binomial random variable)
 N = Total number of observations made
 p = Proportion of workforce that is "working"
   (estimate of the population)
   (also called category proportion)
               p = w / N
 S.D.  = Standard deviation of the sample

 I = Range of error of the population characteristic
 I / 2 = limit of error
 K = Number of standard deviations about the mean
   (established from level of confidence)
  for 90% confidence, K = 1.645
  for 95% confidence, K = 1.96
 Confidence interval = (p - I/2 ) to ( p + I/2)
  where I = 2 * K * S.D.
 Confidence limits = end points of a confidence interval
 Confidence coefficient = relative frequency that the interval will
  contain the population mean

 
 Assumptions of work sampling

Worker activity is assumed to be the same as worker productivity.  This is clearly not always true, but it is true that idle workers will not perform work.  It could easily occur that a worker is classified as "working" when in fact the work is being done wrong or the worker could even be doing something not related to the project (making a project to later take home).

For the statistics to be valid, the observations of each worker must satisfy the criteria of the independence of successive trials.  This means that

(a) The observations must be random.  Every worker must have an equal opportunity of being observed.  There should also be no sequential relationships and the worker must be rated at the instant first seen.  It also means that

(b) The data are collected from the "same" population.  That is, the characteristics of work must be the same.  This is perhaps a major weakness of applying work sampling on construction projects.  As the "mix" of trades varies, it would appear that "percent working" will also change.  Masons laying brick on a simple wall may have a rating as high as 80% while a complex forming task might have a rating as low as 40%.
 

Sample size vs cost

Clearly the more detailed the data, the larger the sample should be.  Management must make this decision.  To a great extent, the sample size will be dictated by the use to be made of the sampling results.
 
 
 Benefits of work sampling

Work sampling is a measure of the effectiveness of management.  In fact, some union leaders will resist work sampling studies if it appears to them that it is designed to be used as a "club" over labor.  Thus, it is important that management undertake these studies with the understanding that shortcomings in the work activity on the project reflect shortcomings in management.

Work sampling results derived from a single tour of a project are not of themselves useful to management.  Work sampling studies are desired to conduct trend analysis studies.  It is only when the work sampling studies indicate that productivity is dropping or going down that management should become concerned.  Work sampling does not tell management what the problem is, but rather that there is a problem.  It is not appropriate to simply compare results of work sampling studies between projects.  Too many differences could occur that would make the comparison between projects meaningless (different observer or sampler, different job task, different phase of construction, different size of project, etc.).

Work sampling results are timely.  No great amount of number crunching is required to yield meaningful results.  The results can be in hand on the same day of making the observations.

Work sampling focuses on labor, the most controllable resource on most construction projects.
 

Labor utilization factor

To many project managers, a simple number as "percent working" is not sufficient to indicate how a project workforce is doing.  To satisfy the need for more detailed information a modification can be made to the traditional method.  This is the labor utilization factor and incorporates three categories (instead of two), namely "effective work", "essential contributory work" and "ineffective work".  This provides a category for the seemingly gray area that often lies between the conventional groups of "working" and "not working".  The purpose is to detect inefficiency more readily.

"Effective work".  This is the process of adding to the unit being constructed such as placing materials in their final location, attaching materials to their final location, moving within 10 feet of the work area, etc.  (Any motion that is essential for progress in developing the project).

"Essential contributory".  This involves activities that do not directly add to the unit being built but work that must be done to accomplish it.  This includes clean-up, receiving instructions, building supports, erecting scaffolding, moving within 10 to 35 feet of the work area, etc.

"Ineffective work".  Doing nothing or doing something that does not add to the unit being built.  It includes walking empty-handed, moving over 35 feet from the work area, using a wrong procedure (difficult to detect the first instant), waiting, etc.  For even more informative reports, the categories of the reasons for the ineffective work might also be stated.  This would give the cause of the delays, such as equipment delay, material delay, crew delay-same, crew delay-other, supervisory delay, misc. delay, personal delay, etc. This is very similar to what Henry Gantt was proposing in one of his idleness expense charts.

Clearly the definitions of the three category designations are open to interpretation.  This is why contractors often set up their own definitions of what should be included in each category.  For example, waiting for a truck may be classified as contributory essential or as not working (ineffective).  This could mean that a second truck is needed.  On the other hand, waiting for a concrete bucket might be classified as "effective".

Let p = the number of workers observed as performing "effective work"
 c = the number of task activities classified as "essential contributory"
 L.U.F.  = Labor Utilization Factor

 L.U.F.  = (p + c/2) / N
or  L.U.F.  = (p + c/4) / N

Note: Differing weights may be given to the essential contributory work classification.

 WORK SAMPLING PROBLEMS

1. Anticipated category proportion = 40% working
 Limit of error = 3%
 Find n if the confidence level = 90%
 
 
 
 
 
 
 

 Find n if the confidence level = 95%
 
 
 
 
 
 
 

2. Anticipated category proportion = 55%
 Confidence level = 90%
 Find n if the limit of error = 2%
 
 
 
 
 
 
 

 Find n if the limit of error = 4%
 3. Measured category proportion = 65% working
 Sample size = 280
 Find the limit of error if the confidence level = 90%
 
 
 
 
 
 
 

 Find the limit of error if the confidence level = 98%
 
 
 
 
 
 
 

4. Measured category proportion = 55% working
 Sample size = 420
 Find the confidence level if the limit of error = 3%
 
 
 
 
 
 
 

 Find the confidence level if the limit of error = 5%
 

 FIVE MINUTE RATINGS

Work sampling studies can be conducted quickly, but they still require the investment of a fair amount of time.  Such studies stand up reasonably well to a critical statistical review.  If the rate of work is below expectations, further analysis will often be required to identify the specific source of the problem.  If an operation is suspect and management wants a little more information, a five-minute rating may be most appropriate.

Five minute ratings are designed to make reasonably quick assessments of the productivity level of a crew performing a specific task.  The results are not designed to be subjected to rigorous statistical analysis.  The data will not be sufficiently detailed to warrant such an examination.  However, since the data is related specifically to a particular operation, management may place some reliability on the results.

Five minute ratings are performed by carefully watching all workers in a crew.  When a study is instigated, each worker involved in an operation is observed for one minute.  At that stage, an assessment is made as to whether each individual worker was "working" for at least 30 seconds or not.  If a worker is determined to be "working" for at least 30 seconds in the first minute, the worker will be noted as having been "working" for the first minute.  This is done for each worker that was observed during this first minute.  The procedure is repeated during the second minute.  This is repeated for each successive minute.  This procedure is followed each successive minute until the number of minutes of observation equal the number of workers in the crew or until the crew has been observed for five minutes, whichever is greater.  At the end of the observation period, the number of "working worker" minutes is divided by the total number of workers multiplied by the number of minutes of observation.  The result is the percent of the workers that are working, a value not unlike that obtained from work sampling.  Note that the percentage working value will not be associated with confidence levels as no statistical rigor would be warranted.  The information is subjective but it can be quite valuable.
 FIVE-MINUTE RATING

Date: _______________________________  Project: ______________________________

Crew Task: _____________________________________________________________________
 

           WORKER DESCRIPTION

Time
 
 
 
 
 
 
 
 

Start
 1
 2
 3
 4
 5
 
 Crew Activity
 
 

Total Worker Units = __________________  Effectives Units: _________________
 
 
 
 

   Crew Effectiveness = ____________________%
 

 DELAY SURVEYS

In recent years (since the late 1970's) more attention has been given to the use of delay surveys.  These delay surveys generate more detailed information than do the work sampling reports.  With work sampling, the "ineffective work" category may give some idea of the source of delays, but the information is not very specific.  Delay surveys have been developed which not only quantify the amount of delay time incurred, but which also isolate the source of the delays.

The most popular form of delay survey is known as the "foreman delay survey" or crew delay survey.  It is a survey that is completed by the foremen or first line supervisors on the project.  Each foreman estimates the total amount of time lost by each crew during each day because of specifically noted sources.  When multiplied by the number of workers in a crew, an idea of the magnitude of the problem is available.

The other delay form is known as the "worker delay survey".  It generates essentially the same information as is generated by the foreman delay survey, but is completed by each of the workers.  This is the least popular of the two survey forms, probably because of the high cost involved in accessing each worker.

A note of caution should be offered concerning the delay surveys.  It is human nature to blame others for your own problems.  Look for consistency of reporting between foremen or workers.  Outliers may have to be excluded from consideration due to the subjectivity of the survey responses.

 FOREMAN DELAY SURVEY
 

DATE OF SURVEY:      CRAFT:

FOREMAN'S NAME:      WORK AREA:

NUMBER OF WORKERS IN CREW:    TYPE OF WORK DONE:
 

 MANHOURS LOST PER DAY

  number of  number of
 CAUSE OF DELAY hours delayed x workers delayed = MANHOURS

 1. Waiting for Materials   x   =
 2. Waiting for Tools   x   =
 3. Rework from Design Errors   x   =
 4. Rework from Field Errors   x   =
 5. Waiting for Instructions   x   =
 6. Waiting for Inspection   x   =
 7. Work Area Not Ready   x   =
 8. Work Area Over-Crowded   x   =
 9. Waiting for Equipment   x   =
 10. Equipment Breakdown   x   =
 11. Delayed by Owner Change   x   =
 12. Interference with    x   =
  Another Crew
 13. Other: _____________   x   =
   __________________
 

COMMENTS:
 
 

 WORKER DELAY SURVEY
 

DATE OF SURVEY:    WORK AREA:

FOREMAN'S NAME:    TYPE WORK DONE TODAY:

JOB CLASSIFICATION:    NUMBER OF HOURS WORKED TODAY:
 

CAUSE OF DELAY    LENGTH OF DELAY (HOURS)

 1. Waiting for Materials
 2. Waiting for Tools
 3. Rework for Design Errors
 4. Rework for Field Errors
 5. Waiting for Instructions
 6. Waiting for an Inspection
 7. Work Area Not Ready
 8. Work Area Over-Crowded
 9. Waiting for Equipment
 10. Equipment Breakdown
 11. Delay by Owner Change
 12. Interference with Another Crew
 13. Other: ______________
   ___________________
 

COMMENTS: