Alcoholism (Part 1)
Building a Control Group

Click to view:

Commentaries on this study

Part 2:  Jung Rules
Part 3:  C.E.O. Carter Rules
Part 4:  Testing Reader Suggestions

Why We Need Control Groups

When doing any research one needs to have a "control" group to compare to the "experimental" group. If you were studying the effects of exercise on heart attacks, you would need to have an experimental group that exercised and a control group that didn't. But that is not enough; you would also have to "control" their diets to make sure that one group wasn't eating more fatty foods than the other group. Similarly in astrological research it is not enough to pick out 100 criminal charts and compare them to 100 non-criminal charts. There are astronomical factors that must taken into consideration.

If one has a group of 100 criminals and 12 of them had Scorpio rising, is this significant or normal? You might just calculate that the likelihood of having Scorpio rising is 8.33% (100 charts /12 signs) but you would be wrong. Due to the tilt of the earth there is not an equal distribution of rising signs at all latitudes. Here is the distribution of rising signs in London courtesy of the British Astrological Association.

Any Londoner is almost three times more likely to have a Scorpio ascendant than a Pisces Ascendant.

Similarly, if you looked at a group of criminals and 75% had Pluto in Leo, this would be meaningless if most of your sample was drawn from the middle of this century when Pluto was in Leo. Not many people know that aspect patterns are not evenly distributed, even beyond the inner planets. Due to astronomical factors Mars is four times more likely to be conjunct the Sun than oppose it. Mars is retrograde 9% of the time and Pluto 33% of the time (See Mark Pottenger's "Astronomical Frequencies" in Astrological Research Methods: Volume I, published by ISAR, pp 203-233). Clearly we need a way to "control" for astronomical effects so we don't come to the wrong conclusion about which counts are significant.

You could use the FAR program developed by Mark Pottenger to calculate the probability of these astrological effects, or use JigSaw to build a control group that mirrors your experimental group closely enough to automatically accommodate these astrological factors. If your control group was from the same latitude as your experimental group, it will automatically tell you what distribution of rising signs to expect. Similarly, if it is from the same range of years it will automatically tell you what distribution of planets by sign to expect.

In this article, I will take you through the considerations that must be made when building a control group for astrological studies. Then I will take you through a step-by-step tutorial on how to use AstroDatabank and JigSaw to build an experimental and control group.

Birth Data Factors to Consider in Building an Astrological Control Group

The primary concern in building an astrological control group is that you control for the non-random aspects of your experimental group. Here are some examples:

• If your experimental group all comes from 1925-1975, so should your control group.
• If your experimental group all comes from the Northern Hemisphere, so should your control group.
• If your experimental group is 90% from the United States, so should your experimental group be.
• If your experimental group is 70% day births, your control group should NOT be 70% day births unless for some reason your selection of the birth data for your experimental group had a bias towards day births.

Did you get caught on the last one? It presents the key point, how random is your selection of birth data from an astronomical point of view? Let's go through the birth factors one at a time:

Year

Your experimental group should have an even distribution of birth years covering a range that is at least equal to the orbital period of the planet in your study with the longest orbit. If it does not, you do not have a random sample of years. Thus you must create a control group that has the same range and distribution of years.

Month and Day

Your experimental group does not have to have an equal distribution of month/days (Sun signs), as long as there was not a month/day bias in the selection of data. If you select 100 musician charts from the AstroDatabank and 50% are Pisces, that is significant and should not be replicated in the control group. We say this because we know that Lois Rodden did not choose, based on Sun sign, which musicians to put in her AstroDatabank.

Time

Similarly your experimental group does not have to have an equal distribution of birth times, as long as there was not a time bias in the selection of data. If you are working with data sets containing at least 60 timed records, that should not be a problem. If you are working with less than 60 records (the rationale will be provided in the next section) or all noon records, than you must match birth times in your control group if you are counting:

• Rising or Midheaven signs and degrees
• Planets in houses
• Planet in Gauquelin Sectors
• Planets aspecting angles
• Rulers of houses
• Parans
• Arabic Parts (Planetary Pictures which use the Ascendant in the formula)
• Sect
• Moon sign and declination
• Moon and Sun degree for Sabian symbols, et. al. and aspects to the Sun and Moon
• or any other astrological features that revolve around time considerations.

Of course, if you are using non charts, none of the above should be counted.

Latitudes and Longitudes

The data in the AstroDatabank is not evenly distributed around the globe. Thus your control group should always match the experimental group on latitude and longitude if you are counting any of the time related factors listed above.

There is also a danger of having your control group too closely mirror your experimental group. If there was a significant Sun sign effect and you chose to replicate the month/day distribution, then it will show up in the control group as well and you will not know there was an effect. If you have a rising sign effect and chose to replicate time, latitude and longitude, then the rising effect will be duplicated in the control group and you won't know there was a rising sign effect – or any of the other time based effects listed above. It is a balancing act. If you think about what astronomical factors you are investigating the choice of what to replicate falls out.

For you Sagittarians, here's the big picture. We are balancing the risk of building a control group that too tightly mirrors the experimental group and masks the effect vs. finding effects that aren't really there because the control group did not mirror the experimental group closely enough (you're measuring noise). (Statisticians call these type I and type II errors.) The first one is like listening to static on the radio and thinking you are hearing words, the second is like listening to words and hearing static. In astrology research the big concern is that we are looking at random patterns and reading patterns into them. Thus we should err on the side of building control groups that more closely mirror the experimental group. If we go too far we won't see what we were looking for, but if we do find something – we know we have something big. It's like looking through a weaker telescope. You won't find things that aren't there, but when you do find something you can be sure it is something important.

How Many Records is Enough for Experimental and Control Groups

It is beyond the scope of this paper to discuss how many records are required in an experimental group to get statistically significant results. But there is a statistical rule of thumb that one should shoot for a theoretical possibility of at least five observations per "box". Thus if one were counting signs, one should use at least 60 charts (5x12 signs). To take a ridiculous example, if you have six musician's charts and half of them have a Pisces Sun this is probably not significant. If one were counting elements one could get away with 20 charts (5x4). JigSaw has a wonderful feature of automatically telling you if you have enough records for statistical significance if you click on any bar chart.

The same minimum requirements apply to control groups, but it is a good idea to go beyond the minimum. If you use a control group generator like the one in JigSaw, it is a relatively easy matter to make large control groups. A rule of thumb for generated control groups is to make them 10x the size of the experimental group. Using a data set this size in the days of hand counting would be quite a burden, but in the computer era it's a cinch.

The larger the data set, the smaller the likelihood of getting spurious results. It is harder to get exactly 500 heads in 1000 tosses than five heads in ten tosses.

Step-by-Step Directions for Building an Experimental and Control Group

For the purpose of this example we will go through the steps to build an experimental and a control group for alcoholics using AstroDatabank and JigSaw. There are 264 alcoholics in the AstroDatabank with a Rodden Rating of B or above. We must use a Rodden rating of B or above if we are to test house positions in an astrological study of alcoholism.

There are three major steps in building a control group:

1. Examine the distribution of years, latitudes and longitudes in the experimental data. (Since we have many more than 60 records in our experimental group we will not examine the distribution of times.)
2. Filter out the outliers so you get a more uniform set.
3. Generate a random set of birth data to match the same distribution of years, latitudes and longitudes.

Step 1: Examining the Distributions

We observe the following distributions of years, latitudes, and longitudes:

Step 2: Filtering Out the Outliers

The distribution of pre-1900 years is not as even as it is after 1900, so we will only use data after 1900. Most of the data is from within the United States, so we only use the data between 25N and 50N and 71W and 124W.

We get 213 cases when we filter for:

• Alcohol Abuse
• Rodden Rating B and above
• Date >01/01/1900
• Latitude between 25N and 50N
• Longitude between 71W and 124W

(click to enlarge)

All seemed regularly distributed across the United States. (From a purely statistical point of view, whatever results we get from this research will only be applicable to drunks in the US. One can only draw conclusions based on the characteristics of one's sample.)

Step 3: Generating a Random Set of Birth Data to Match the Experimental Group

I used JigSaw to build my control group. Since I have over 60 records, I can choose the easiest method of building a control group - using data replication. Here are the steps:

1. From the AstroDatabank File menu, choose Export, select JigSaw, then type in the file name "ADBAlc.DAT."
2. From JigSaw's File menu, choose Open and Select "ADBAlc.DAT."
3. From the JigSaw Research Menu, select Generate Test Data.
4. Click on Use Current Project as Base.
5. Click on Replicate Years and Replicate Lats & Longs.
6. Press OK.
7. Type in AlcCtrl for a filename and press Save.

That's it. You just made a control groups of 213 records that have random dates and times and the same distribution of years and locations as the experimental group. In general, the replication method provides the best control group. Its distribution of years and places exactly matches that of the experimental group. Any anomalies in the experimental group are perfectly mirrored in the control group so the chances of getting false signals are reduced.

There is only one thing more that can be done to improve the control group. A bigger control group is more likely to represent the population than a small one, which by chance might end up having some abnormalities. Just as is harder to get 500 heads in 1000 tosses, than five heads in ten tosses, so too you are less likely to randomly get too many Scorpio Moons with a larger data set.

We can increase the Alcoholic control group to over a thousand records by repeating steps 2-8 five times. Give the files names: AlcCtrl1, AlcCtrl2, AlcCtrl3, AlcCtrl4, and AlcCtrl5. Back in AstroDatabank, make a new file, import each one and calculate the data. You now have a control group of 1065 records that mirror the year and place distribution of the experimental group.

JigSaw is a popular Windows research program available through Astrology Software Shop. It is terrific for exploratory work using graphs.

Next week we will use them to begin exploring the astrological signature for alcoholism using Jungian criteria.

Note:

There is a slight problem with this control group. About 10% of these random cases would likely have abused alcohol in their lives. This will "water down" the differences we see between the known alcoholics and the control group by about 10%.

"A community study conducted in the United States from 1980 to 1985 using DSM III criteria found that about 8% of the adult population has Alcohol Dependence and about 5% had Alcohol Abuse at some time in their lives…A United States national probability sample of non-institutionalized adults (ages 15-54) conducted in 1900-1991 using DSM-III-R criteria reported that around 14% had Alcohol Dependence at some time in their lives, with approximately 7% having Dependence in the past year."

p. 202 Diagnostic and Statistical Manual of Mental Disorders Fourth Edition Published by the American Psychiatric Association Washington, DC Copyright 1994

by Mark McDonough

Don't miss
our Links page
for the most extensive
annotated list
of astrology research
articles and resources
on the web.

Help Us Build an AstroSignature for Alcoholism

You've seen how we've built AstroSignatures for Excess Water and Excess Fire, and for Planets Above and Below the Horizon. Now help us build an AstroSignature for Alcoholism. Help us find the astrological traits we should look for in alcoholics. Give each of your suggested rules a rule weight, as we have in these previous AstroSignatures. We'll plow in your ideas as we continue creating this AstroSignature.

Come back next week to check out the comments that others have left.

AstroDatabank offers this privately-funded forum for astrologers who want to share astrological insights in a respectful and educational way. AstroDatabank does not condone or support comments that are profane, obscene or discriminatory against anyone’s race, color, religious creed, national origin, ancestry, gender, disability, sexual orientation, veteran status or any other similar category. While we make an effort to remove as soon as we can any comments that we deem inappropriate for these or other discretionary reasons, we cannot guarantee that we are always able to do so in a timely way given the nature of the Internet. However, we reserve the right to remove comments or to block participation at our sole discretion and at any time. Please understand that users’ comments are their own and do not reflect the views, techniques or interpretations of anyone at AstroDatabank. We hope that you will use this forum to share your astrological insights with your colleagues in a respectful manner and that you will enjoy reading others’ comments on the chart under discussion. For more information, please read our legal disclaimer and terms of service.

Where do you want to go now?

• See an index of AstroSignatures
• Learn more about how AstroDatabank works
• Return to the home page

2/10/2000: