The new (with version 4.4) Family Finder Report (FFR) of the Family Tree Maker (FTM) product allows searching for any portion of your FTM file against the whole collection of FTM CD volumes, FTM User Home Pages, and other WWW sites selected by FTM. The seeker can freely use the Home Pages and other WWW sites, but the FTM CDs must be purchased in order to garner whatever information they may contain.

There have been numerous complaints about the FFR, particularly the use of married names when searching for women. I decided that it was time to run some controlled experiments to determine if there really was a problem.

Conclusions

The FFR searching algorithm is flawed to the point of rendering it nearly useless. The justification offered by Broderbund for including female married names in the searching is that this was done to improve the utility of the tool. The intent is to find more hits, especially in cases where the female spouse appears in the data under her married name rather than her maiden name. It is unfortunate that the effect is exactly the opposite of that desired: So many spurious results are returned that it becomes a practical impossibility to make any use of them at all.

I managed to find a moderate-sized family on one of the World Family Tree (WFT) CD-ROM volumes which does not seem to be duplicated elsewhere. This was the Van Vl¤ck family of 71 people on WFT 04.2715. From this family, I selected two individuals, both unmarried, without offspring, and with known birth and death dates—one male and one female.

The nature of the experiments were to run FFR against test families with varying parameters and measure how effectively the target pedigree, WFT 04.2715, was found. All experiments were performed on the morning of Saturday, 30 May 1998.

The male subject (Martin Van Vleck) was isolated in a fresh workspace and married to an invented Margaret Smith with plausible birth/death dates. They were given four children: two males and two females, with plausible birth dates (year only). A FFR was run against the whole family of six.

Results

Twenty-nine pages of listings, with the desired pedigree, 04.2715, listed first.

Experiment 2

The female subject (Margaret Van Vleck) was isolated in a fresh workspace and married to an invented Henry Smith with plausible birth/death dates. They were given four children as in Experiment 1, and an FFR was run against the whole family.

Results

91 pages of listings. The listings for the subject start on page 32, with the desired 04.2715 on page 43. Peeking at the listings for the subject, they all seem to be for Margaret Smith, her married name. It was difficult to tell the proportions of hits on her married name vs. her maiden name as this information is concealed and must be revealed individually for each entry in the list. Every one I tried at random (~20) was for Smith.

Experiment 2a

This was identical to experiment 2, except the invented husband was "Martin NoName".

Results

A list of 6 entries, with the target 04.2715 first. Three of the remaining entries were for FTM CD volumes that also appeared to be plausible hits, but could not be checked because they were not in the experimenter's possession.

Experiment 2b

Identical to experiment 2, except that the FFR was restricted to the female subject only. The family was left intact.

Results

30 pages of listings, all for the subject. The target 04.2715 was on page 20. Again, all the hits seem to be for the subject's married name.

Experiment 2c

Identical to 2b except the subject was severed from the invented spouse.

Results

Results identical to experiment 2a.

Experiment 1c

The male subject of experiment 1 was separated from his invented spouse and the FFR was run for him alone (as with experiment 2c)

Results

A list of two items, the first of which was the target 04.2715. The second was for a user's home page that could not be checked ("Page has no data").

Experiment 3

A FFR was run for the whole Van Vl¤ck family of 71 individuals.

Results

Thirty pages of results. The first two entries were the only ones with more than 10 indicated individuals, the rest indicating one or two individuals for the most part. These first two are the two found in experiment 1c.

Experiment 3 proves that the chosen subjects are well isolated and do not appear in more places than the one chosen WFT (04.2715). The second result of experiment 3 could not be checked due, apparently, to a stale link. Assuming this a bona fide finding, the appearance of the subjects in two sources is still a reasonable isolation.

Comparing experiment 1 with experiment 2 shows how FTM's search algorithm greatly favors finding male family members over female family members. These two experiments were symmetrical, with the roles of male and female reversed from one experiment to the other. In experiment 1 the male subject was easily found, but in experiment 2, the female subject's "hit" was lost in the middle of nearly a hundred pages of spurious results.

Experiments 2a, 2b and 2c were run to further understand the effect seen in experiments 1 and 2. Experiment 2a, where the female subject was married to a spouse with a unique surname (NoName), shows the disastrous effect of searching with the woman's married name: 91 pages (.02% success) vs. 6 lines (16.67% success).

Experiment 2b was intended to show the effect of the husband's unique surname compared to having no husband at all. The result was surprising. Although the search was restricted to the female subject, the fact of her extant marriage to Smith continued to confuse the results.

Experiment 2c shows that a husband with a unique surname is equivalent to no husband at all in the FFR search algorithm. For comparison, experiment 1c was run, with the male subject isolated from his invented spouse, with similar results to experiment 2c.

Software

Explanation of headings in Table
 

Experiment	The experiment number, as referenced in the text.
Subject	The name of the primary subject.
Spouse	Spouse of the subject. These are invented people.
Results	The number of results listed. (Only one is valid.)
Success Rate	Reciprocal of the number of results, expressed as a percent success.
Target Location	Approximate position of the target, 04.2715, in the results. A position at or near the top is desired.

-fin- © 1998 Roderic A. Davis, 2nd All Rights Reserved »HOME

Email: dav4is@yahoo.com Snail: Roderic A. Davis, 2nd P.O. Box 118 Hyde Park, NY 12538 USA