Problems building a survey with architect : Roster names; linking a categorical bin stage with a specific roster name question

Hi !

I´m finding some technical problems while I´m building my survey with Architect.

First of all, I need to map several relationships with a (always the same) roster name generator and I notice that if one name is called/selected by the interviewed participant once, than this name is not appearing in the name list in the other questions of the survey. I tried to fix this problem adding a different roster (different csv) for each question (prompt) of the survey (even if it is the same list of names). However, this strategy is not working and the names are still removed from the name list as they are called along the survey. Someone knows how I can fix this?

Then, for some relationships (eg. Who is giving you technical advice), I would like to categorize the names that are called by the participant from the roster (name list) in function of the modalities of the technical advice (eg. A.Giving technical advice orally / technical advice in practice). I managed to do that creating first a Name generator for roster data stage, and then a categorical bin stage (linking the stages with a rule). However, the only way to have just the names called/listed by the interviewed participant selectively/specifically from the question that is interesting me, is by creating a new “node type". Otherwise, in the case that I´m using the same “node type” as the other questions, I will have the cumulated list of the names that have been selected/called by the participant all along the survey (and therefore the names are not strictly associated with the question of the technical advice). Any idea of how I could fix this problem without creating a new node?

Thank you in advance for your kind help!

Best regards,

Hi Merlin,

This sounds like interesting research and I appreciate you working with Network Canvas for it. We have discussed this question among some of the team. While I might have an answer it’s based on some assumptions about your work, so if these aren’t correct then get back to me and we can adapt.

The first assumption is that you need to use the roster name generator interface rather than a quick add interface with a roster as a side panel. Admittedly the roster name generator is nicer to look at, sorts, has searching, etc…So it’s generally a nicer way to see the names from a roster. However, what it does not have is a side panel for ‘nodes already nominated’. That means that if you want to use this interface it should generally be for any questions that exclusively remove people from the roster of “potential alters” into the pool of nominated alters.

So this means that to use this roster screen on multiple screens one after the other should include the preamble “is there anyone whom you have not nominated that…”. If you want to draw from both a roster and a list of people already nominated on the same screen then you should use a name generator ‘form’ or a name generator ‘quick add’ screen and include the roster and ‘people already nominated’ in side panels.

Also, for each one of these interfaces when you drag the nodes to the nomination drop zone on the right it just adds the node to the alter pool for that node type. It does not assign additional variables by default. It sounds like you would want to have a variable for each node called ‘gives technical advice’ or technical_advice. Then you can set that to True for anyone you have nominated via the first roster. Then later, you might want to do something like a sociogram screen or a categorical box to resolve the ones who were nominated on such a screen.

So for example, imagine I nominate some people who give technical_advice from a roster. Then later I’m asked to nominate people from recall (who may also give technical advice, but we aren’t sure). You can for example, either filter down to those who have not reported True (or False) on technical_advice and then sort them into a categorical bin, do an alter form with a single True/False checkbox or arrange all the nodes on a sociogram with technical_advice = True highlighted. In the latter case you can then say “please tap on all nodes who give technical advice, some nodes may already be highlighted from earlier questions”.

Then with the technical_advice question now checked against everyone who was nominated (i.e the entire alter pool), you can then filter to technical_advice = True for a question that drills down into the type of technical advice.

Might that help your conundrum? If not, feel free to elaborate and I or a member of the team will see what we can do to help further.

Hi Bernieh,

Thank you so much for this deep and useful reply!!

Learning to work with Network Canvas is really interesting; I´m happy to learn to use this powerful tool.

I divide my questions in 2 parts :
A.roster issue,
B.Nodes (categorical bin) issue.

Thank you in advance for your availability and your help!

Best regards,

A.Roster issue

I confirm (after a first pilot study), I need to use a roster of names mainly because it is very useful as:
1.the names in this cultural context are (sometimes) hard to spell for me,
2.I´m doing a (kind of) whole network study and I´ve already the list of all the names,
3.the roster name in Network Canvas is really intuitive.

My concern is that if I use the same name list (roster file in CSV format), when a name is mentioned in one relationship (i.e. let´s say Matt is mentioned for the question 1), it doesn´t appear anymore for the next relationship (i.e. Matt is not an option anymore for the question 2). However, for the study, it is important that Matt continues to be an option for all the relationships I would like to map…

Intuitively, what I did in a tentative to fix this is that I changed one letter of one variable (ie. The column “FamilyName”) adding 1 letter for each roster needed (i.e. For the 2nd relationship that I want to map, my CSV file column name is “Familynamee”, for the 3rd it is “Familynameee”). However, I suppose there must be another way to do that. Moreover, this strategy might be a problem for the database management after the data collection. What do you think?

B.Node (categorical bin) issue

Concerning the question of the nodes, the research questions only have to do with one node (smallholders), in which some give more technical advice (=one variable of the same roster), some are more conflictive (= another variable of the same roster), etc… (I want to map a decade of relationships). Therefore, my questions are:

If I create an “artificial” node with network canvas architect, this will not be a problem for the data analysis?

In case you recommend me to avoid creating an artificial node, how could I create a categorical bin (i.e. modalities of technical advice: .oral / .in practice) specific for one network (i.e. technical advice)? I would like to see only the names mentioned by the participant in this specific network (i.e. farmers who give me technical advice) in order to add them in one of the categories (.oral / .in practice).

In case you confirm that if I create an artificial node type (i.e. technical advice) it will not affect the analytical part of the study, then I think that I understood how to create a categorical bin specifically associated to a node type (considering your recommendations).

I think the important thing to grasp is that the “mental model” for the roster name generator is that you are adding participants to the interview network, and not nominating them as having an attribute.

We have discussed implementing a special mode in the software that would essentially automatically import all roster members into a network when the interview begins. You could then use the existing attribute nomination interfaces to add the attributes to only specific network members. However this feature is not yet implemented.

One possible workaround would be to do this manually: have an initial interview stage where you add all roster members to the interview, and then follow that stage up with name generator/interpreter stages that add the attributes you need.

This is a very broad question! It will require you to think about your data differently, certainly, which may or may not fit with the way you are used to working.

This can be accomplished by first setting an attribute based on the node having given some form of advice (for example, setting gives_advice = true on a name generator prompt using the additional variables feature - see the documentation), and then subsequently implementing a categorical bin interface that uses the stage level network filtering feature to only show nodes that have this attribute.