Sample selection methods

To use the fixed interval selection method, you specify:

• The interval value that Analytics generates when you calculate the sample size
• A start number greater than zero and less than or equal to the interval value

The start number and the interval value are used to select which records are contained in the sample.

Considerations

When you use the fixed interval selection method, you need to be alert to any patterns in the data. Because a fixed interval is used for sample selection, a non-representative sample can result if the data has a pattern that coincides with the interval you specify.

For example, you sample expenses using an interval of $10,000, and the same expense category appears at ten-thousand-dollar intervals in the file, which results in all the selected records coming from a single expense category. This type of scenario is uncommon, but you should be aware that it could occur.

To use the cell selection method, you specify:

• The interval value that Analytics generates when you calculate the sample size
• A seed value used to initialize the random number generator in Analytics

The interval value dictates the size of each cell. The random number generator specifies which monetary unit or which record number is selected from each cell.

The seed value

If you specify a seed value, it can be any number. Each unique seed value results in a different set of random numbers. If you respecify the same seed value, the same set of random numbers is generated. Explicitly specify a seed value, and save it, if you want to replicate a particular sample selection.

Considerations

The main advantage of the cell selection method over the fixed interval selection method is that it avoids problems related to patterns in the data.

For monetary unit sampling, two disadvantages exist:

• Amounts can span the dividing point between two cells, which means they could be selected twice, yielding a less consistent sample than the sample generated by the fixed interval method.
• Larger amounts that are less than the top stratum cutoff have a slightly reduced chance of being selected.

To use the random selection method, you specify:

• The sample size, as calculated by Analytics – that is, the number of samples to select
• A seed value used to initialize the random number generator in Analytics
• The population size – that is, the absolute value of the sample field, or the total number of records in the data set

For classical variables sampling, sample size and population size can be automatically prefilled by Analytics.

The random number generator specifies which monetary units or which record numbers are selected from the data set. Each selection is a random distance apart.

The seed value

If you specify a seed value, it can be any number. For classical variables sampling, the seed value must be a positive number not greater than 2,147,483,647.

Each unique seed value results in a different set of random numbers. If you respecify the same seed value, the same set of random numbers is generated. Explicitly specify a seed value, and save it, if you want to replicate a particular sample selection. You can also retrieve a seed value from the command log.

Considerations

Large amounts may be excluded from a monetary unit sample

With the random selection method, each monetary unit has an equal chance of selection, and there is no guarantee that the resulting sample will be evenly distributed. As a result, the distance or gap between selected units may be large in some instances. If all the monetary units associated with a large amount happen to fall into a gap, the amount is not included in the sample. There is also no top stratum cutoff available when using the random selection method.

With the fixed interval and cell selection methods, there is an assurance that the selected units are evenly distributed, or relatively evenly distributed. And top stratum cutoff is available.

Amounts may be included more than once in a monetary unit sample

Analytics does not generate the same random number twice, however random numbers that are close, or sequential, can occur.

With monetary unit sampling, close or sequential random numbers equate to close or sequential monetary units being selected, which in turn can lead to an associated amount being selected more than once.

With record sampling and classical variables sampling, the same problem does not exist because each random number equates to a different record.

Random number algorithms

For record sampling and monetary unit sampling, the random number generator in Analytics has two algorithm options:

• Mersenne-Twister
• The default Analytics algorithm

Mersenne-Twister is a widely used random number algorithm and it has better statistical properties than the default Analytics algorithm. Use the default algorithm if you require backward compatibility with Analytics scripts or sampling results created prior to Analytics version 12.

For classical variables sampling, Mersenne-Twister is not an option and the default Analytics algorithm is used.

1. In the source table, create a computed field that uses the following expression:

   RECNO( )

   For more information, see Define a conditional computed field.

2. When you sample the data, output by Fields, not by Record.

   You must output by Fields in order to convert the computed record number field to a physical field that preserves the record numbers from the source table.

3. Include the computed record number field in the output fields you specify.