SAMP for Windows

SAMP for Windows (Samp 1.2)

Surrey European Management School
University of Surrey, Guildford GU2 5XH
Phone: +44 (0)1483 259347
Fax: +44(0)1483 259511
email: Sems@Surrey.ac.uk
UK: £49.50; course licence £99
US: $79; course licence $149
RoW: £70; course licence £120
Minimum hardware requirements: Windows 3.x

SAMP, the pioneer sampling simulator, is now - Good Heavens! - nineteen years old. To celebrate, it has been given cosmetic surgery, i.e. a Windows interface.

All the familiar features remain. Almost two decades later the indecisive simulated citizens of hypothetical Little Stripling-Under-Wey (population 11,041) have yet to resolve the question of whether to redevelop their town centre. As further procrastination they commission a sample survey to estimate the mean of a single variable, the u-shaped distribution of attitude toward redevelopment.

SAMP users may provide an estimate for them by drawing:

a simple random sample of up to 10 percent,
a clustered sample using 20 enumeration districts 'designed to be somewhat homogeneous in attitude'. One chooses the number of clusters and then the total sampling proportion.
a stratified sample using three strata varying in marketing research SES levels. (Favourability is inverse to status.) One chooses the overall sampling proportion, which is then applied to each stratum. Results appear for the three strata and the total sample.
a (hiss!) quota sample. One chooses the total number of males, females, young, middle, and old and gets results with a built in bias (too few upmarket and thus anti- development respondents in the middle age category.)

For each simulation the user receives means, medians, standard deviations, etc. A separate command gets the true result. Promise not to tell? The population mean is 3.172 with a standard deviation of 1.654.

The program is accompanied by extensive on-line Help information including the equivalent of a chapter on sampling in an introductory text.

All this has been translated from DOS to Windows by John McLoughlin, building on Nigel Gilbert's original DOS version. I am not much of a Windows user but those who like stabbing at trompe-l'oeil plastic knobs should be totally satisfied.

I'm afraid, however, I have to note some ragged edges: 1) The Further Reading help section is blank. 2) While the simulations report medians and the text stresses their instability in a u-shaped sample, the true median is not reported. 3) The program can not be installed unless you remember that the letters in the serial number must be caps. 4) There is a typo in the quota sample instructions. They read 'Quota sampling...is based on random sampling. Instead...' rather than 'Quota sampling...is NOT based on random sampling. Instead...'

While SAMP's appearance has caught up with the fashions of the times, its mental abilities have not. To be blunt, it doesn't do a heck of a lot, considering the price and complexity.

It is far from flexible. (1) The only adjustment one can make in the parameters is to change N. There is no way to look at parent distributions other than the cemented-in, u-shaped attitude. (2) In stratified sampling each stratum must have the same N, which makes it impossible to show how one can improve precision by shifting cases from solid to iffy subgroups. (3) Users can store only one sampling for each method.

More telling, SAMP has not become more ambitious, just more attractive. Let me cite three areas.

First, it prints cost-effectiveness numbers such as call backs, travel time, and interviewer pay. But these are just boiler plate print statements. I am shocked, shocked I say, that a Management School would not build in routines to simulate the likely cost of various sampling schemes.

Second, to keep things simple, the user's decision is almost always a sampling proportion, not a sample N. This vitiates the pedagogical principle that, in practice, N is crucial, the sampling fraction unimportant.

Third, and most important it would be extremely laborious to simulate sampling distributions, i.e. the results of repeated samplings. The program is not especially fast (on my low end, 33MHz 486 PC it takes 15 to 20 seconds to draw an SRS sample of 550 cases). For an individual student to draw the hundreds of samples necessary for a simulation is not practical. On the other hand the computer, with its extraordinary immunity to boredom, is itching to do just that. (The diskette accompanying Richard Maisel and Caroline Hodges Persell's How Sampling Works [Pine Forge Press, 1996] simulates sampling distributions for a variety of sampling plans and parent distributions.) To me the key concept in teaching sampling is sampling distribution, not sample, and thus, handsome as it is, SAMP misses the point. Furthermore, in the absence of replications, comparisons among methods can be quite deceptive because quotas might well do better than probability samples on a single draw.

SAMP is a landmark in the application of computers to methods teaching. The scenario gives a realism to sampling that is missing from coloured beads and poker chips. And obviously, the computational power of computers allows for more realistic Ns than the usual class room demonstration. It is good news that SAMP is available in a popular format. Nevertheless, in my judgment it needs more than cosmetic surgery lest it become the Dorian Gray of academic software.

James A. Davis
NORC/University of Chicago