The Master Network

Before considering the statistics of the Master Network, the visual and group data provides a lot of information.  In reference to Figure 3, the Advancing Group (labelled 1) clearly dominates the network in size and connections.    Notice how the larger bundles of edges generally travel between the Advancing Group and the rest of the network in Figure 2.  This image shows a different view with the edges unbundled to allow some more visibility of intergroup edges compared to the Advancing Group.  The expanse of this group is seen here where the edges for all nodes in Group 1 are highlighted:

Notice how the red eclipses the gray on the graph.  The Advancing Group reaches further in the network than any of the other groups.  Part of this makes sense given the size of the group compared to the others.  The Advancing Group includes 536 nodes, more than double the size of any other group.  For perspective, 536 nodes of the 1,460 total means Group 1 includes about 36.7% of the network's people.  This matters for the discussion on the name of the group and the process of the trials.


Figure 2 displays all the edges without the nodes.  The highlighted red section appears the darkest grey in that image which emphasizes the connectedness of the Advancing Group.  Other groups show strong interconnectivity, but none to the same extent.  This all suggests the importance of the individuals in Group 1.

Other noticeable patterns that are recognizable just by looking at Figures 1, 2, and 3 are the group formations.  Several of the large groups, noticeably groups 1, 2, 3, 4, 5, 6, and 7 have a denser center with peripheral nodes.  This offers observations about how the social nature of the trials played out.  The clusters reflect different aspects of the trials, discussed further below, but the pattern emerging in these groups shows certain nodes control smaller networks.  These tight knit groups in the Advancing and 1693 Trials groups interact more with each other, but form the same patterns as the Bradbury and Nurse and Procter Groups.  Central people or smaller groups within each cluster connect, like their own mini-network.  The role of this pattern is explained more below with the example of the Dorcas Hoar Group.

The Master Network and Eigenvector Centrality

Who are the members of the Advancing Group?  Most of the highest Eigenvector recipients are located in this cluster.  Note that Eigenvector Centrality refers to a node's power based on its position in the network.  Someone like John Indian, a slave, may hold a strong position in the network, but his social power never escapes his bondage.  The skew of the power favors a limited number of people.  The total of all the Eigenvector scores equals 1, and rounded to the thousandths place, only 383 nodes, or 26.2% of the network, receive a score of .001 or higher.  Below is a frequency graph to show how extreme the Eigenvector Centrality slant in the Master Network is to the top 26%.

The Y axis represents how frequently the Eigenvector score on the X axis occurs in the network.  The X axis Eigenvector scores are rounded to the thousandths place, but the repeating numbers stand for intervals within the thousandths.  Intervals with a zero frequency were removed to clear the graph.  The numbers within the graph are the labels of the exact frequency, and in several cases that a frequency occurs in that position given the difficulty to show a frequency of 1 compared to 938.  The far right of the graph is the maximum score, where John Hathrone's node is represented.  The far left represents the vast majority of nodes in the network.  There are so many people to consider in the discussion on the trials, so to narrow it down, this will only cover the most important people.  The following graph compares the Eigenvector scores of the nodes equal to or greater than half of the maximum Eigenvector score, .013.

These 15 people represent 1.02% of the total network population, a quite literal 1%.  The sum total of their Eigenvector scores is .145.  Since the total Eigenvector Centrality scores total 1, these 15 people receive 14.5% of the centrality.  These 15 share a disproportionate influence on the network, and all of them are found in the Advancing Group.  The Salem Witch Trials happen as a result of these people's actions and involvement.  The Advancing Group takes its name since the people sorted into it were responsible for the trials.  The top 15 people represent three major categories-

1. Officials: Hathorne, John; Corwin, Jonathan; Herrick, George Gedney, Bartholomew

2. Accusers: Putnam, Ann Jr.; Walcott, Mary; Hubbard, Elizabeth; Lewis, Mercy; Williams, Abigail; Warren, Mary; Bibber, Sarah; Sheldon, Susannah

3. Supporters: Parris, Samuel; Putnam, Thomas Jr.; Ingersoll, Nathaniel

Hathorne, Corwin, and Gedney served as judges for many of the examinations and trials, so nearly every case involved one or more of these men.  Herrick, as a marshal, arrested many suspects and summoned witnesses.  Although Herrick rarely appears as a major figure in histories of the trials, this speaks to the use of social network analysis.  Herrick only testified in a few instances, but he often stood in the court room to escort the accused suspects to the stand.  Few historians waste any ink on him, according to their indexes, Mary Beth Norton refers to him only four times and Baker and Boyer & Nissenbaum have zero references.  This disconnect seems unusual since even documentary evidence exhibits his involvement.  In December 1692, he petitioned the colony for assistance since for nine months, "in Serving of Warrants and Apprehending many prisoners, attending Examinations & Courts of Oyer and Terminer,...have often conveyed Prisoners unto Prison and from Prison to Prison, it hath taken my Whole time and made me incapable to get anything for the maintenance of my Poor family."  He spent so much time serving the county for the witch trials that he lost all opportunity to work and sufficiently provide for his family.  Traditional history methods miss this since in comparison to the verbal testimony of suspects, the brief references to Herrick bringing a suspect into the court room or signing off on the execution of an arrest warrant are easily overlooked for more exciting topics.

The accused suspects appear in this influential group since everything occurs in the context of their words.  However, none of these girls are from Andover, the town where the most accusations took place.  They lived in Salem Village or Salem Town, and aside from Bibber, age 36 in 1692, the girls were ages 11-20.  Their accusations led to the most arrests, and their accusations shaped how the trials advanced despite other groups of accusers and towns joining the hysteria.  While their number of accusations is not always greater than other accusers, their accusations targeted more important suspects earlier on which crafted the trajectory of the trials.  Here is a list showing the number of accusations made by each of these women.

Putnam, Ann Jr.: 83 accusations

Walcott, Mary: 77 accusations

Hubbard, Elizabeth: 54 accusations

Lewis, Mercy: 60 accusations

Williams, Abigail: 62 accusations

Warren, Mary: 60 accusations

Bibber, Sarah: 20 accusations

Sheldon, Susannah: 40 accusations

Despite the differences in accusations, the people they accuse make them powerful in the network.  These accusers form the usual group in the cases, and although their involvement changed over time.  Bibber made far fewer accusations than Susannah Sheldon, but her early involvement in the examinations in March 1692 connected her to more significant figures.  Sheldon started to accuse later, so a few well connected figures were less involved for Sheldon to form a relationships.  When Bibber accused Sarah Good and Martha Corey early in their cases, Bibber attended the examination.  Sheldon often accused suspects after the original claims, starting to name witches in late April, and she became more active as Bibber, who accused more in March, accused less people.  A person's influence on the network depended on not only who they knew, but when they started to involve themselves, which is why Andover names do not appear in the top Eigenvector scores of the network.  In fact, the first Andover name in the Eigenvector ranking is number 52, the confessor Richard Carrier with a score of 0.004.

The third category, supporters, are three men without whom the trials never would have happened.  Parris's daughter and niece, Betty and Abigail respectively, and Putnam's daughter and maid, Ann and Mercy respectively, fell victim to witchcraft early on in the trials.  Ingersoll not only joined Parris and Putnam in filing complaints and reporting as a witness of the afflictions, but his tavern held many examinations when the crowds watching overwhelmed the meeting house space.  This allows Parris, Putnam, and Ingersoll to connect with other accusers and the accused throughout the entire hysteria.  Parris even served as a scribe for several examinations which provided him the opportunity to write the official record for Hathorne and Corwin to review later.

The Master Network and Betweenness Centrality

Much like the Eigenvector scores, the Betweenness Centrality scores skew to a small portion of the graph, but with a significant difference of who ranks the highest.  Below the table shows the top 10 Betweenness Centrality scores.

A few names repeat from the Eigenvector list, which is for the same reason they appeared there.  Hathorne, Warren, Putnam, Walcott, and Corwin played a role in a large number of cases compared to everyone else, so they understandably provide pathways to a lot of the network.  The one notable difference between the lists is Warren's position.  Since Mary Warren accused suspects before facing accusations and confessing, she bridges a gap between accusers and accused.  Her position connects her to both groups more than other accusers by embracing this role.  This shows how even the number of relationships does not always matter in connectivity to the graph.  Warren has 180 degrees compared to Putnam's 211.  Even with 30 degrees less, Warren connects into the network better than Putnam through her blurred line between victim and perpetrator of the accusations.

Stoughton and Sewall appear on this list as important agents of the Court of Oyer and Terminer.  Stoughton served as the chief judge of the court and Sewall worked as the clerk.  This meant Stoughton absolutely attended most trials.  Sewall prepared many documents for the court cases and his name appears on many of the warrants which demonstrated his role in the background of the events of 1692.  While more time could be spent on the dynamics of the court, the accused suspects provide the more fascinating insights in this section.

Consider Group 8 in Figure 3, Dorcas Hoar's Group.  The central node of this wheel and spoke pattern is Hoar.  Her case file is small and distinct enough to form its own group without the clutter of petitioners around it.  The pattern for her is the same pattern for Mary Bradbury, Rebecca Nurse, and Elizabeth Howe although their groups involve more people and Howe's case is in the Topsfield Group.  The accused suspect sits in the center.  Those testifying against the suspect extend outward from the center, sometimes to other spokes to form triangles around Hoar.  Each clearly distinct section that extends outward is from specific documents referring to the same testimony or testimonies from the same incident.  This is how the witch trials functioned.  Neighbors testified against specific individuals with limited connections elsewhere in the network, even in the same case file.  The accused connected to the accusers in the Advancing Group.  This is where Group 1's name comes from, accusations advanced through the trials as this group took up the accusation.  Hoar's group shows how small clusters formed independently, but needed this massive group to cause the witch hunt against a person.

However, as the Bradbury Group, the Nurse/Procter Group, and the Topsfield Group show, some of these smaller clusters grew to a significant size as well.  In the case of Elizabeth Howe, it was more the number of neighbors involved the distinguished her.  In Bradbury and Nurse's cases, neighbors came out to defend these women.  The petitions on the behalf of Bradbury and Nurse add a lot of people to the network, but look at their groups.  The petitioners connect in the same wheel and spoke pattern as anyone testifying against the suspects.  These positions lack power in the network.  Those defending suspects never gained a significant position compared to members of the Advancing Group.  If their only involvement was signing their name to defend Bradbury or Nurse, it made little impact.  The accusing girls held too much power.  The Court of Oyer and Terminer convicted both Bradbury and Nurse.  Rebecca Nurse and Elizabeth Howe hanged on July 19, 1692.  Mary Bradbury survived when Gov. Phips intervened in the trials before her execution, and she was eventually granted a reprieve.  In never mattered in the trials who testified for and against a suspect.  Hoar was also convicted but survived for the same reasons as Bradbury.  However, Nurse, Bradbury, Howe, and Hoar were expected to suffer the same fate despite different types of connections in the network.