Collecting data can be time consuming and expensive, therefore there should be a clearly defined purpose for data collection. In addition, the following questions may help:
What data do you think you will need? Will sampling be needed? How and who will collect the data? Do they understand the purpose and have they
been trained? What will the data look like once it has been collected? Create dummy tables. Will it tell you something about the process that you want to improve? If the data
do not bring you to an actionable step, you may want to rethink the data collection.
Sampling is an important concept for process improvement. If you must collect data, be sure that you are clear about what data will be most useful. The following section on sampling provides some guidance.
Definitions: A sample is a collection of numbers (measurements or counts) on the quality characteristic (QC) of process variable (PV) of units drawn from a process. A sample unit is a part of the process on which the QC or PV is measured. The units may be patients, incidents, time (day of week), a form, an incoming shipment of material--all of which are integrally related to process.
There are two methods for selecting sample units that will be introduced in this course:
Stratification can be done with both these methods.
Which patients would you pick to measure waiting time for this physician's office?
Definition: A data collection method designed to select sample units in a block of a predetermined size.
Purpose: Quota sampling is used to select sample units so as to capture the detailed behavior of the process. A quota sample is especially useful for situations where the data is time or sequence dependent. This data collection method is likely to capture the detailed behavior of the process.
Guidelines:
The sample units must be in sequence and serially numbered.
 |
How to select a quota sample:
Suppose you want to collect numbers (data) on
n
units in the process(
n
;is the sample size--for example, if data on 25 patients is collected then
n
is equal to 25). Select a location and time that you wish to start collecting data. Select the first unit in the process at that time and location and every unit that
follows until you have data on n
units.
Which patients would you pick to measure waiting time for this physician's office?
Definition: A data collection method designed to select at fixed or count intervals.
Purpose: The method is used to select sample units over extended periods of time.
Guidelines:
The sample units must be in sequence and serially numbered.
 |
Systematic sampling is especially useful in situations when you want to:
Sample over extended periods of time. Reduce the impact of a variety of conditions to prevent systematic errors from
influencing the analysis of the data.
If the team wanted to show the impact that time of day and day of week has on QC variability, they could use a systematic sample. For example, a group might elect to study the elapsed times from registration to being seen by a physician. They could measure the first patient to walk in at two hour intervals. Alternatively, they could collect data for every 5th patient over a two week period.
How to select a systematic sample:
Example:
- You want to survey 25 patients for a run chart.
n = 25- The clinic sees 100 patients per week.
T = 100 T/n = k; 100 divided by 25 = 4.
k = 4- Sampling Plan: You will interview every
4th patient for 1 week.
Definition: Knowledge of the process allows you to group things so that you get the most information from your data collection for the effort expended. This is called stratification. Creative thinking is usually required to select possible groupings, or strata , within which sample units will be taken. Examples of possible strata are:
Acuity
Day of Week
Zip code
Sex
Age
Weight
Purpose: Allows you to better understand the variation in your QC or PV by aligning data with known groups in the process.
Guidelines:
The sample units must be in sequence and serially numbered. The grouping variable must be selected based on process knowledge. There are two types of stratification:
Pre-stratification: This occurs when the team breaks the data collection into smaller groups prior to collecting data, because they know the grouping. In pre-stratified samples, you plan to collect enough data in the strata to establish a run chart for that strata.
Post-stratification: This occurs when the team collects additional information during data collection (such as triage level). They then separate the data into groups based on this additional information after data collection is complete. In post stratified samples, there is usually not enough data in the strata to establish a run chart and, hence, further data collection may be necessary.
In both types of stratification, the data are analyzed within the
strata
or groupings. Prior to data collection, stratify to the maximum extent that you can based on your
knowledge of the process(
pre-stratification ). If you are unsure of the impact of some potential stratifying factor, plan to collect
the data necessary to post-stratify on it. Dr. Deming in his works on sampling observed that it may not be possible or even wise to
examine every strata. His advice was to judgmentally select a few strata that are
representative of the process under the
worst
conditions and--resources permitting--under the
best
conditions. By understanding the behavior of the
process with respect to the QC or PV under these extremes, it is usually possible to
improve the process over the entire range of conditions encountered.
