Changes in Personal Social Networks across Individuals Leaving Their Street Gang: Just What Are Youth Leaving Behind?

Round 1

Reviewer 1 Report

Short wrote (1960, The Sociocultural Context of Delinquency) wrote that “attributes, such as sharing a gang name, do not explain gang behavior. In order to explain gang behavior we need to understand relationships and social interactions.”

This paper has done more interesting research than a majority of gang research papers, but it commits the error Short described.  The analysis and interpretation confuse the relations and attributes. Gang research has over decades confused relations and attributes:  three or more people who commit crime, is probably the worst definition one can conjure. Gang names are attributes. Social ties among people who report the same attribute tells us nothing about social networks.  While the paper uses the term, personal social network, the analysis is not a network analysis.  The paper uses terms such as gang, engaged. disengaged, membership and so on, which perpetuate ambiguity. 

Here's an example of the source of confusion Short tried to get beyond in gang research. 

"They described the gang “exit” process as moving through stages: (1) first doubts, where gang members contemplate the symbolic and instrumental value of their current role; (2) weighing alternative roles, where gang members engage in anticipatory socialization of new or different roles; (3) turning points, which function as crystallization of discontent
to act on the aforementioned considerations; and (4) post exit certification, which works to validate new roles while inoculating gang members from old ones."

The paper does not include personal network analysis measures. There's no structural analysis, which network analysis brought into social science. For example, centrality measures are necessary to understand group structure.  How does exiting affect group centrality?  Gangs are groups but what differentiates 12 boys on a corner and from 12 boys who are gang members on a corner?  "Gang exit process" can be measured and exiting should alter group structure. 

Before publication, the author should rethink the analysis and, first, look at the data as either attributes and relations.  "Post exit certification, which works to validate new roles while inoculating gang members from old ones" includes ambiguous, confusing terms and social processes, which network analysis can help measure, or not.  Perhaps post exit certification means a change in attribute. Does certification have network measures? probably not, but what does it mean, how can it be measured. 

Gang research is packed with word salad. Example: "This finding clearly reemphasizes the growing agreement that disengagement is an important process to be studied and that, not surprisingly, crime desistance is more clearly tied to full disengagement than de-identification as a member of a gang."  How can these ambiguous notions be translated into network measures.  There's no citation for the Connect Survey.  It seems, too, that network methodology has borrow from publish network studies but without citation.

PLEASE also see attachment

Comments for author File: Comments.pdf

Author Response

We have uploaded a new version of the paper in tracked changes. Thank you so much, Reviewer 1, for your comments pointing out the importance of clarifying and citing key terms and to be careful not to be ambiguous. We have revised the manuscript to make each term more clear and cite the relevant papers that follow the bulk of the current literature. More details on our revisions are below:

Key comment 1: "The analysis and interpretation confuse the relations and attributes. Gang research has over decades confused relations and attributes:  three or more people who commit crime, is probably the worst definition one can conjure. Gang names are attributes. Social ties among people who report the same attribute tells us nothing about social networks.  While the paper uses the term, personal social network, the analysis is not a network analysis."

We have better explained the large and growing body of network analysis literature that includes ego-centric analyses to indicate that the measures used in the paper are network measures. We expand the discussion of ego-networks in the introduction and add additional cites for ego network measures as network analysis throughout the text. Personal network composition using measures of ego-alter ties has been defined as network measures by well-known social network analysis scholars (Steve Borgatti and Tom Valente) for the last few decades, although is less known to crime scholars than socio-centric analyses. Borgatti is an author on a book titled “Egocentric Network Analysis” (Perry, Pescosolido, Borgatti, 2018), making note that network composition is a key part of the social network analytical framework. See also Halgin and Borgatti 2012. An Introduction to Personal Network Analysis and Tie Churn Using E-Net (Connections. Volume 32, 37-48). Halgin and Borgatti’s analytical approach involves a number of measures defined as compositional but they indicate, they are nonetheless network measures and is typical of measures derived from a personal network research design or PNRD, as was the data collection method in the current study. To provide more details to the reviewer, Valente (2010) writes that (p.61): “personal network exposure is the number or proportion of ties holding a particular belief or engaging in a particular behavior.”

We have clarified these measures in the current paper by adding the term “personal network exposure” where appropriate and citing Valente. Characteristics of ties and measures of strength of ties fall under “relational” network measures; those that solely describe a node (or in our case an alter’s characteristics) are “attributes.” Aspects of attributes describing ego-alter ties are used to create network measures, that are distinct from structural measures created from alter-alter ties. We have made that clear. Furthermore, Valente (and others) describe key “network” measures within the network framework as those representing strength of relationships and types of roles and relations. Table 4-1 in Valente’s book uses the term “compositional measures” to be part of egocentric network analytical measures, as we do in the current paper. This may be confusing to readers because measures related to composition such as “proportion of relations that ego goes to for advice” is compositional, but it is also a network measure because it describes roles and/or the relationship between the ego and alter. It is not an attribute in the sense that it only describes the node. Basically, in egocentric analysis, attributes of relations are network measures.

For further clarification if needed, we have uploaded in the attachments a table from Crossley, Bellotti, Edwards, Everett, Koskinen and Tranmer (2015), page 77 showing that these ego-alter attributes of relations are network measures distinguished from structural.

We are open to suggestions from the reviewer to further identify additional ways we can clarify that our paper, as it is written and now further clarified, focuses on network relational measures.

Key comment 2: There's no structural analysis, which network analysis brought into social science. For example, centrality measures are necessary to understand group structure.  How does exiting affect group centrality?  Gangs are groups but what differentiates 12 boys on a corner and from 12 boys who are gang members on a corner?  "Gang exit process" can be measured and exiting should alter group structure. 

We have further clarified that our paper did not intend to include the structural network measures that are typical of socio-centric analysis. We also address this in the further research section.

Key comment 3:  Related to Decker and colleagues' role transition framework: "Post exit certification, which works to validate new roles while inoculating gang members from old ones" includes ambiguous, confusing terms and social processes, which network analysis can help measure, or not.  Perhaps post exit certification means a change in attribute. Does certification have network measures? probably not, but what does it mean, how can it be measured. "

We have clarified on page 5 that the study's intent was not to assess their framework and in the limitations section we discuss that our study's temporal period was too short to assess certification.

Key Comment 4:

"Gang research is packed with word salad. Example: "This finding clearly reemphasizes the growing agreement that disengagement is an important process to be studied and that, not surprisingly, crime desistance is more clearly tied to full disengagement than de-identification as a member of a gang."  How can these ambiguous notions be translated into network measures?"

This particular comment really that helped us see that we needed to be much more careful how and when we used the term "disengagement" especially because we were later defining disengagement as those respondents who cut off all interaction with their former gang members. We have reduced the use of the word "disengagement" and replaced it with more precise language. For instance, in the introduction (pages 1 and 2), we deleted disengagement and replaced it with "leaving the gang."

We also replaced the language in the “Other Group Processes” category of measures with the phrase “Other Group Characteristics.” These measures here were only used for descriptive purposes for the baseline characteristics table and not in analyses.

Comments 5 and 6. "There's no citation for the Connect Survey.  It seems, too, that network methodology has borrowed from publish network studies but without citation."  

We have added a citation for the Connect Survey and added a number of network analysis citations in the measures section and earlier paragraphs (Valente, Blau, etc.). We  also added an additional network measure and statistic related to tie dispersion and derived by Blau (1977). The measure is described in the measures section and in Tables 3 and 8. We believes this is helps to show the variety of meaningful measures related to ego networks.

Reviewer 2 Report

Dear author,

This was an excellently written and interesting article to read. I believe it will make a significant contribution to the field, especially (as you highlight) the current lack of research surrounding ego networks. I only have a couple of minor suggestions to make for revision.

1) Line 226, you mention stage 2. Please expand on role exit theory, mentioning the other stages.

2) Line 241, this is a very long sentence that’s difficult to read, consider shortening this or breaking it up.

3) Line 314, you mention the Eurogang criteria. Briefly outline the key components of a gang, as per this criteria. Also, line 333 you say that they don’t have to meet the self-reported definition of the eurogang criteria, which seems to contradict line 314. Please consider clarifying this (i.e., mention difficulties in self-reporting membership, so you rely on outreach workers classifications)

4) For tables 2-4, avoid repeating information presented in tables within the text too. 

5) In your 2nd to last paragraph, you discuss cure violence. It would be interesting if you add some of the findings regarding effectiveness to date

As I say, a really interesting and well needed paper.

Author Response

Thank you so much for helping to make the paper more readable. We have uploaded a new version of the paper with tracked changes. We believe this revision is streamlined and more clear in many places.

1) Line 226, you mention stage 2. Please expand on role exit theory, mentioning the other stages.

Response, we have re-written the sections of the narrative where we mention role exit theory to make it more clear that we are referring to Decker, Pyrooz and Moule's discussion of role transitions and role exit. These changes can be found on pages 5 and 6. Then in the conclusion, we indicate that it is unfortunate we cannot test their framework or compare our findings to theirs.

2) Line 241, this is a very long sentence that’s difficult to read, consider shortening this or breaking it up.

We have broken up the sentence into two sentences.

3) Line 314, you mention the Eurogang criteria. Briefly outline the key components of a gang, as per this criteria. Also, line 333 you say that they don’t have to meet the self-reported definition of the Eurogang criteria, which seems to contradict line 314. Please consider clarifying this (i.e., mention difficulties in self-reporting membership, so you rely on outreach workers classifications)

We have clarified our enrollment criteria and the Eurogang definition components on pages 7 and 8.

4) For tables 2-4, avoid repeating information presented in tables within the text too. 
We have shortened this section to only include highlights of the statistics shown in the tables.

5) In your 2nd to last paragraph, you discuss cure violence. It would be interesting if you add some of the findings regarding effectiveness to date.

We have added a few sentences describing this program as generally effective when implemented with fidelity.

Round 2

Reviewer 1 Report

Researchers gathered composition data on "people people know."  Composition data have limitations on interpretation.  Engagement requires structural measures.  Disengagement requires structural measures. Example:  "Characteristics of the respondent’s personal networks (p. 11)" requires structural measures. The paper offers an extensive review of gang and network literature; that's great. But absent of relational data, there's little to  say about engagement and disengagement and friendship, which are relational concepts.  Person 1 hangs out with person 2.  Hangs out requires structural measures.  The interchageable use of the concepts group adds ambiguity, absent of a definition of gang v. group.  All groups are not gangs.  Gang members commit crime but that begs the question what are structural differences in social groups v. gangs that help explain the difference between gangs and groups.  Are there structural differences in a cohort of 20 youth whose members commit crime vs. 20 youth who don't commit crime.  Attribute data doesn't explain behavior.  Jargon adds misinterpretation: engagement requires centrality measures; so does disengagement.  Meat Head the gang member commits crime in a gang but not in a group.  An attribute might help explain his behavior but attributes alone can't explain his change in behavior; structural measures, centrality, betweenness.  Gangs v. groups might have different measures of density.  Friend v non-friend require structural measures.  Researchers cited McCarty (2002): his was a study of network structure.  Valente (2010) looked a network exposure re criminal behavior.  Exposure requires structural measures. A network of youth, say, 20, can act to alter behavior, crime v no-crime, but structural measures help explain the behavior.  Age differences linked to crime help understand behavior, 25 year old v. 15, but structural relationship help explain behavior change.  Jargon adds confusion, engagement v disengagement.  attributes alone won't explain this behavior.  Structural measures offer a starting point for interpretation.

Author Response

Thank you for your additional comments about the paper and the revisions. We provide further documentation below, documenting that the paper, as developed and analyzed, focuses on important aspects of desistance, as mentioned by scholars. Similarly, as documented in the previous response, the included measures are core measures of social networks. Structural measures are only one component of measuring networks. We have added these citations to the revised manuscript.

Decker & Lauritsen 2002: Stated that persistent gang ties are an important factor associated with desistance from gangs and further research study any differences between persistence in ties versus more sudden elimination of ties. 

• Studied the path involved with leaving gangs.
  • Either suddenly –  (Elder 1998)
  • Or more gradually - Bushway, Piquero, Broidy, Cauffman, & Mazerolle, 2001;

The following paper documents the importance of describing network changes related to composition. They also document that describing interactions such as hanging out or and doing things with relations constitutes behavior, which fits in nicely to what Decker and Lauritsen suggested above:

Feld, S. L., Suitor, J. J., & Hoegh, J. G. (2007). Describing Changes in Personal Networks over Time. Field Methods, 19(2), 218–236. https://doi.org/10.1177/1525822X06299134

They write: "Social ties are continuously being created and lost as well as changing their nature over time. We emphasize that network descriptions are specific to their particular definitions of ties. Then, we suggest that studies of change can focus on: (1) individual ties, or whole personal networks; and (2) whether ties are gained or lost, or change their characteristics over time." They also state on page 220 that behavior includes interactions: “As time marches on, new interactions and feelings are accumulated, the older ones become less relevant, and the characteristics of the relationship change. Thus, it is reasonable to think of the usual characteristics of a relationship as moving averages of the thoughts, feelings, and interactions that the reporting individual has experienced over some indeterminate time period.”

Furthermore, Paul Bellair, a well-known scholars published a piece documenting the importance of describing networks using compositional measures:

Bellair, P. E., Light, R., & Sutton, J. (2019). Prisoners’ Personal Networks in the Months Preceding Prison: A Descriptive Portrayal. International Journal of Offender Therapy and Comparative Criminology, 63(3), 383–405. https://doi.org/10.1177/0306624X18799575