How the Program Works
Assume that an archaeological survey has been conducted, and that a number of sites have been located. Since real-world surveys always produce a sample of the sites that existed in the past, based partly on survival of the sites and partly on our survey methods, the actual site universe from the past cultural system will be somewhat larger than the number of sites found. Archaeologists realize that even in cases of full-coverage surveys, we are still taking a sample of a larger population.
The RankSize program was written to accommodate this fact. The user supplies a value from .01 to .99 which represents the estimated site recovery rate. In cases where such estimates are not known, a figure which represents the survey coverage of an area may be used (and probably adjusted downward somewhat to account for the non-survival of some sites). Thus, the RankSize program requires the user to address issues related to sample estimation. When the user specifies, for example, that the site recovery rate is 50%, what is implied is that the site universe actually contained twice as many sites as those recovered. The program calculates the number of sites in the site universe by dividing the number of observed sites (which may be up to 400) by the sample fraction, rounding to the nearest whole number if necessary. For the sake of archaeological reasonableness, assume that the largest site in the settlement system has been located.
These two pieces of information (the number of sites in the site universe and the size of the largest site) become the basis of the simulation, in that they describe the parameters of a hypothetical site universe. The RankSize program constructs an ideal, log-normal population of sites based on these parameters. Then, a series of random runs is performed on this population. In each run a number of sites equal to the sample proportion is chosen randomly (without replacement) from the log-normal site population. The program calculates the value of the Kolomogorov-Smirnof (K-) statistic between the drawn sample of sites and the ideal, log-normal expected site distribution (with the same number of sites as in the sample) and compares it to the value of the actual sites, accumulating the number of times the value of K- from the simulation is greater than or equal to the value from the real sites. The process is repeated for up to 65,000 runs.
Besides running the Monte-Carlo simulation, you can also use the Rank-Size program to make on-screen graphs, and copy them to graphics or word processor programs. You can adjust different values, including the slope of the "expected site distribution" and the cutoff point at the lower end of the site distribution. You can also calculate the least-squares regression of the slope of the observed sites, and then run the simulation again to see if the regression line describes the observed sites better than the log-normal distribution does (the regression does not accommodate residuals, so often the log-normal line fits the data better than the regression line). You can also use the "MultiPlot Tab to do quick plots of up to four different sets of sites at the same time. This routine does not run simulations, or allow you to change values, but I wrote it to deal with the problem of having to put several rank-size lines on one figure for articles.
What the Program Does
Essentially, the program asks the question, "If the site universe from
which the observed sample was drawn actually conformed to a log-normal
distribution, what is the likelihood of drawing a sample from this population
with a K- value greater than or equal to that observed in the real data?" 
If the values from the simulation exceed those in the real data a sufficient
number of times, it can be inferred that the observed sites probably do
not deviate very much from a log-normal distribution, and the greater deviations
can be caused by simple sampling error. The program reports the percentage
of runs in which the simulation values exceed the observed value, and this
number may be used as a rough estimate of probability of obtaining a difference
as large or larger than the observed, by chance selection from a true log-normal
population. Rather than supply a simple "pass/fail" answer at a pre-arranged
alpha level, the use of a probability figure allows more equivocal results
to be judged (see Savage, 1997; Falconer and Savage, 1995; abstracts from
these articles are on my Abstracts Page).
Program Details
The original RankSize program was written in Turbo Pascal to run in versions of DOS after 3.3. Though not a Windows program, it can be configured to run in a window in Windows 3.1, Windows NT, or Windows 95. The program manual that comes with the software explains how to set it up for your computer. The manual is in the file "RankSize.man" and it is in WordPerfect 5.1 format.
The RankSize Version 3.2 program has been rewritten in Visual Basis for Windows. The screen shots shown above are from this newer version. It contains on-line help and is considerable easier to use than the old DOS version. You can download either version below. Version 3.2 differs from version 3.1 only in the way the chart control operates, and in the way the installation program runs. Setup programs created on Windows XP platforms try to overwrite the Visual Basic runtime files on earlier platforms, and do not succeed. Nothing that Microsoft tells you to do for this problem actually fixes it, so I've taken matters into my own hands and done the VB runtime installation in a different way that requires a different setup program. Version 3.2 should install and operate on any Windows operating system from version 98 onward.
Downloading the RankSize program:
| Windows Version (RankSize 3.2) | DOS Version (RankSize v2.04) |
|
Click on the icon below to download the Windows version of the program. You'll get a Zip file called "RankSize3.2.zip" which you'll need to unzip to a scratch directory. Having done that, run the SetupRankSize.bat program. This batch file first installs the Visual Basic runtime files, then proceeds to the RankSize program. I recommend that you accept the default locations for files.  Get RankSize 3.2 |
The Version 2.04 ZIP file includes the manual, two programs ("Ranksize.exe"
- the actual RankSize program, and "Rsplot.exe" - plot
utility program, which will let you plot up to four sets of data on the
same graph). 
Get RankSize 2.04 |
