[Rsiena-help] What does this error mean?
Tom Snijders
Tom.Snijders at nuffield.ox.ac.uk
Thu Mar 22 19:47:16 CET 2012
Dear Jennifer,
I had misunderstood and thought that only 22% were present at all waves. 57% present in all waves is a lot better.
For analysing only one network and no other dependent variables, conditional is the default, so you have to request non-conditional estimation explicitly in the call of sienaModelCreate.
For composition change, actors can enter or leave the network at any time, at or continuously between the observation moments. When they are present they are included in the simulations; see Huisman & Snijders (2003) (and Snijders et al., Social Networks 2010, for the general explanation of the simulation model used in Siena). This means that an actor entering at the very last time point is meaningless because there is zero time left for taking part in the network.
Best wishes,
Tom
================================================================
Tom A.B. Snijders
Professor of Statistics in the Social Sciences
Department of Politics and Department of Statistics
Nuffield College
University of Oxford
tel. +44-01865-278599
-----Original Message-----
From: Victor, Jennifer Nicoll [mailto:jnvictor at pitt.edu]
Sent: 22 March 2012 17:36
To: Tom Snijders; rsiena-help at r-forge.wu-wien.ac.at
Subject: RE: [Rsiena-help] What does this error mean?
Thanks for your suggestions, Tom and Ruth.
Tom, as I understand the manual, the default is for non-conditional estimation, yes? So, if I have not set a variable for conditional estimation then the estimate I ran is non-conditional. Please correct me if I've misunderstood.
In general, my data can be thought of as a semi-panel. There are three waves, or points of observation corresponding to three different Congresses, and there are some overlapping actors in each wave. The breakdown is approximately:
57% of actors appear in all three waves
22% of actors appear only in one of the three waves 11% appear only in waves 1&2 10% appear only in waves 2&3
Ruth's comment that the data must be continuous and allow for entry and exit from the network *between* waves is an element of the program I had not understood before. If Siena is designed only for such continuous data and not for data of the semi-panel type that I have described, then Siena is indeed not the right platform. Can one of the authors confirm this for me, before I turn away from this modeling approach all together?
Best,
Jennifer
___________________________________________
Jennifer Nicoll Victor
Assistant Professor
Department of Political Science
University of Pittsburgh
4600 Wesley W. Posvar Hall
(412) 624-7204
E-mail: jnvictor at pitt.edu
Homepage: http://www.polisci.pitt.edu/person/jennifer-nicoll-victor
-----Original Message-----
From: Tom Snijders [mailto:Tom.Snijders at nuffield.ox.ac.uk]
Sent: Thursday, March 22, 2012 8:29 AM
To: rsiena-help at r-forge.wu-wien.ac.at
Cc: Victor, Jennifer Nicoll
Subject: RE: [Rsiena-help] What does this error mean?
Dear Jennifer,
If you have only 22% overlap between waves then Siena is unlikely to be a good method for you. The error "unlikely to terminate this epoch" may be a problem especially for conditional estimation, and this problem might be resolved if you change to unconditional estimation (also see Section 14.1 of the Siena manual).
Best wishes,
Tom
================================================================
Tom A.B. Snijders
Professor of Statistics in the Social Sciences Department of Politics and Department of Statistics Nuffield College University of Oxford tel. +44-01865-278599
-----Original Message-----
From: rsiena-help-bounces at r-forge.wu-wien.ac.at [mailto:rsiena-help-bounces at r-forge.wu-wien.ac.at] On Behalf Of Ruth Ripley
Sent: 21 March 2012 20:53
To: Victor, Jennifer Nicoll
Cc: rsiena-help at r-forge.wu-wien.ac.at
Subject: Re: [Rsiena-help] What does this error mean?
Jennifer,
RSiena is modelling a continuous process which you have observed at specific time points. If your actors cannot enter between waves 2 and 3 there is no way RSiena can create any links at time 3 for them.
The manual is showing you how to create edge lists (which are not used in RSiena) via sparse matrices which are. It looks to me as though temp1 is the matrix you need.
Regards,
Ruth
On 21/03/2012 20:43, Victor, Jennifer Nicoll wrote:
> Dear Ruth,
> Across the three waves 22% of observations appear only in one wave.
> There is no opportunity for actors to enter between waves. Does this
> suggest Siena is not a good platform for these data?
> My adjcommxxx objects are class matrix. I created them by following
> the example in the manual, namely:
> #111TH COMMITTEES
> temp1 <- as(comm111.full, "dgTMatrix")
> edges111 <- cbind(temp1 at i+1, temp1 at j+1, temp1 at x) #CREATE EMPTY
> ADJACENCY MATRIX
> adjcomm111 <- matrix(0, 555, 555)
> #PUT EDGE VALUES IN DESIRED PLACES
> adjcomm111[edges111[, 1:2]] <- edges111[,3] #CHECK LENGTH
> length(which(comm111.full !=adjcomm111)) #THIS RETURNS 0, WHICH IS
> CORRECT Best, Jennifer ___________________________________________
> Jennifer Nicoll Victor
> Assistant Professor
> Department of Political Science
> University of Pittsburgh
> 4600 Wesley W. Posvar Hall
> (412) 624-7204
> E-mail: jnvictor at pitt.edu
> Homepage: http://www.polisci.pitt.edu/person/jennifer-nicoll-victor
> -----Original Message-----
> From: Ruth Ripley [mailto:ruth at stats.ox.ac.uk]
> Sent: Wednesday, March 21, 2012 4:29 PM
> To: Victor, Jennifer Nicoll
> Cc: rsiena-help at r-forge.wu-wien.ac.at
> Subject: Re: [Rsiena-help] What does this error mean?
> Dear Jennifer,
> 10's will work correctly in RSiena, but they will not work in sna.
> There is little point having actors who are only present at time 3. If
> you assume they arrive at time 2.5 they could be used in the simulation.
> Similarly you might want to assume that actor 3 leaves at some time
> after 2.0.
> What format are your adjcommxxx? Try class(adjcommxxx) to find out.
> They should be dgTMatrix.
> Regards,
> Ruth
> On 21/03/2012 20:14, Victor, Jennifer Nicoll wrote:
> > Thanks, Ruth. As indicated in the manual, I have the structural
> zeroes in RSiena data set to "10," rather than "NA." This is correct, yes?
> >
> > Could my trouble have to do with the exogenous events file? I
> specified this file as the manual instructed and then used the command:
> > compChange.us<-sienaCompositionChange(usexevents)
> >
> > Where my "usexevents" file looks like this:
> > V2 V3
> > 1 1 3
> > 2 1 3
> > 3 1 2
> > 4 3 3
> >
> > (for the first four actors, where actors 1& 2 are present in all
> three time periods, and actor 2 is present only in the first two time
> periods, and actor 4 is present only in the 3 time period).
> >
> > I received no errors with my sienaCompositionChange command, but I
> wonder if there is a problem?
> >
> > As an aside, I also have trouble with the command:
> > uscom<-varDyadCovar(list(adjcomm110, adjcomm111)) > Error in
> varDyadCovar(list(adjcomm110, adjcomm111)) :
> > not a list of sparse triples matrices > > My "adjcommxxx" files
> are sparse matrices, but I'm not certain about the "triples" part.
> >
> > JNV
> > ___________________________________________
> > Jennifer Nicoll Victor
> > Assistant Professor
> > Department of Political Science
> > University of Pittsburgh
> > 4600 Wesley W. Posvar Hall
> > (412) 624-7204
> > E-mail: jnvictor at pitt.edu
> > Homepage: http://www.polisci.pitt.edu/person/jennifer-nicoll-victor
> >
> >
> > -----Original Message-----
> > From: Ruth Ripley [mailto:ruth at stats.ox.ac.uk] > Sent: Wednesday,
> March 21, 2012 3:14 PM > To: Victor, Jennifer Nicoll > Cc:
> rsiena-help at r-forge.wu-wien.ac.at > Subject: Re: [Rsiena-help] What
> does this error mean?
> >
> > Dear Jennifer,
> >
> > The error you have indicates that the simulation process is failing.
> > Either steps are happening so frequently that the cumulative time
> is > increasing too slowly, or, more likely here, in conditional
> estimation, > every step is choosing to do nothing so the target cannot be reached.
> >
> > One point about sna: having seen some of your data, I wonder if you
> had > any structural zeros in the matrix you used with sna. sna does
> not > recognise these.
> >
> > Regards,
> >
> > Ruth
> >
> > On 21/03/2012 15:38, Victor, Jennifer Nicoll wrote:
> >> Thank you to Ruth Ripley and Nate Doogan for providing help with
> >> this > modeling problem. Based on their suggestions I have
> simplified the model > and the effects; however, I still am unable to
> estimate the model. I'm > seeking some general advice about fitting
> my data. Ruth wondered if > Siena was the appropriate model for these
> data and I'm attempting to > investigate this question and get more
> feedback. I appreciate further > feedback from users of this list.
> I've also posted to the general > Statnet list and have estimated
> ERGMs with these data.
> >>
> >> I have encountered numerous problems relating to degeneracy and
> >> model > specification. The data are from three waves (congresses)
> of US members > of Congress and their co-membership in legislative
> organizations. I have > a number of dyadic and nodal attribute
> variables as well about the > actors. The hypothesis I seek to test
> is whether there are bridging-ties > among the legislators through
> their memberships in legislative > organizations. I suspect that the
> high degree of density in the data may > be part of the problem.
> >>
> >> One approach I've tried is to simply calculate betweenness (in the
> >> SNA > package) for each network of legislators in each time
> period, add this > as a vertex attribute, and estimate an ERGM, in
> which betweenness is a > nodal-covariate; however these models have
> difficulty with convergence > and fit. I am unable to obtain MCMC
> standard errors, and even after > numerous attempts at
> respecification the diagnostics suggest that fit > may be
> problematic.
> >>
> >> Another approach is to use Siena. Siena should allow me to take >
> advantage of the longitudinal nature of the data. But the output that
> > I've been able to obtain suggests that the model is problematic, as
> Ruth > and Nate have pointed out. See output below.
> >>
> >> So my various attempts at testing this hypothesis using these >>
> advanced > methods have left me frustrated. While we can (and have)
> tested the idea > using a variety of less sophisticated methods, it
> seems to me that the > advanced statistical methods, such as those
> available through ERGM and > Siena, are designed for exactly this
> type of modeling; however, I have > not been able to achieve
> satisfactory findings that are reportable. I > can continue to try to
> reparameterize the models to search for fit and > to avoid
> degeneracy, but I would like to hear thoughts of readers of > this
> list regarding:
> >> - the appropriate method for testing the hypothesis - best
> practices >> for iterative modeling and specification for these >
> types of models (given that each iteration is so computationally
> expensive)
> >> - which architecture (ERGM, Siena, something else?) seems most >
> appropriate for the data?
> >>
> >> Siena verbose output:
> >>> myeffus
> >> effectName include fix test initialValue parm >> 1 constant
> caucuses rate (period 1) TRUE FALSE FALSE 100.00000 0 >> 2 constant
> caucuses rate (period 2) TRUE FALSE FALSE 70.41182 0 >> 3 degree
> (density) TRUE FALSE FALSE 0.03424 0 >> 4 betweenness TRUE FALSE
> FALSE 0.00000 0 >> 5 same usparty TRUE FALSE FALSE 0.00000 0 >>>
> myusmodel<-sienaModelCreate(useStdInits=FALSE, projname='siena_us_02')
> >>> usans02<-siena07(myusmodel, data=myusdata,effects=myeffus,
> batch=TRUE, verbose=TRUE) >> >> Stochastic approximation algorithm.
> >> Initial value for gain parameter = 0.2.
> >> Start of the algorithm.
> >> Observed function values are
> >> 1. 123349.0000 2. 5156392.0000 3. 123580.0000 >> >> Start phase
> 0 >> theta: 0.0342 0.0000 0.0000 >> Current parameter values:
> >> 0.03424475 0.00000000 0.00000000
> >>
> >> Start phase 1
> >> Phase 1 Iteration 1 Progress: 0%
> >> Phase 1 Iteration 2 Progress: 0%
> >> Phase 1 Iteration 3 Progress: 0%
> >> Phase 1 Iteration 4 Progress: 0%
> >> Phase 1 Iteration 5 Progress: 0%
> >> Phase 1 Iteration 10 Progress: 0%
> >> Phase 1 Iteration 15 Progress: 1%
> >> Time per iteration in phase 1 = 50.7582 >> Average deviations NR
> generated statistics and targets >> after phase 1:
> >> -39637.750000
> >> 6787139.500000
> >> -39249.250000
> >>
> >> Diagonal values of derivative matrix :
> >> 32772.3901 524963179.7438 38364.4002 >> dfra :
> >> 32772.39 3087677.74 17924.11
> >> 5095442.50 524963179.74 3357687.99 >> 39393.49 3556965.56
> 38364.40 >> >> inverse of dfra :
> >> 0.00029660366044 -0.00000197940840 0.00003466438151 >>
> -0.00000228738847 0.00000001994553 -0.00000067696765 >>
> -0.00009248392723 0.00000018324891 0.00005323684804 >> >> Full
> Quasi-Newton-Raphson step after phase 1:
> >> 1. 26.551774
> >> 2. -0.252611
> >> 3. -2.820084
> >> This step is multiplied by the factor 0.10000.
> >> Intervention 1.4.2: jump after phase 1 decreased by factor
> 26.5517736817607 .
> >> Phase 1 achieved after 16 iterations.
> >> theta: 1.03424 -0.00951 -0.10621
> >> Current parameter values:
> >> 1.034244749 -0.009513884 -0.106210769 >> >> Phase 2 has 4
> subphases.
> >> Each subphase can be repeated up to 4 times >> >> Start phase
> 2.1 >> Phase 2 Subphase 1 Iteration 1 Progress: 3% >> Phase 2
> Subphase 1 Iteration 2 Progress: 3% >> theta 1.0433 -0.0096 -0.0982
> >> ac 1.009 0.985 1.014 >> Phase 2 Subphase 1 Iteration 3 Progress:
> 3% >> Phase 2 Subphase 1 Iteration 4 Progress: 3% >> theta 1.06134
> -0.00977 -0.08233 >> ac 1.006 0.992 1.008 >> Phase 2 Subphase 1
> Iteration 5 Progress: 3% >> Phase 2 Subphase 1 Iteration 6 Progress:
> 3% >> theta 1.07940 -0.00994 -0.06651 >> ac 1.01 0.99 1.01 >> Phase
> 2 Subphase 1 Iteration 7 Progress: 3% >> Phase 2 Subphase 1 Iteration
> 8 Progress: 3% >> theta 1.0975 -0.0101 -0.0508 >> ac 1.004 0.995
> 1.007 >> Phase 2 Subphase 1 Iteration 9 Progress: 3% >> Phase 2
> Subphase 1 Iteration 10 Progress: 3% >> theta 1.1155 -0.0103 -0.0353
> >> ac 1.004 0.996 1.006 >> Phase 2 Subphase 1 Iteration 20 Progress:
> 3% >> theta 1.2058 -0.0112 0.0411 >> ac 1.002 0.999 1.006 >> Phase
> 2 Subphase 1 Iteration 30 Progress: 4% >> theta 1.2961 -0.0121 0.1146
> >> ac 1.00 1.00 1.01 >> Phase 2 Subphase 1 Iteration 40 Progress: 4%
> >> theta 1.386 -0.013 0.185 >> ac 1.00 1.00 1.01 >> Phase 2 Subphase
> 1 Iteration 50 Progress: 5% >> theta 1.4767 -0.0139 0.2526 >> ac
> 1.00 1.00 1.01 >> Phase 2 Subphase 1 Iteration 60 Progress: 5% >>
> theta 1.5670 -0.0148 0.3167 >> ac 1.00 1.00 1.01 >> Phase 2 Subphase
> 1 Iteration 70 Progress: 5% >> theta 1.6574 -0.0157 0.3777 >> ac
> 1.00 1.00 1.01 >> Phase 2 Subphase 1 Iteration 80 Progress: 6% >>
> theta 1.7477 -0.0166 0.4350 >> ac 1.00 1.00 1.01 >> Phase 2 Subphase
> 1 Iteration 90 Progress: 6% >> theta 1.8380 -0.0174 0.4888 >> ac
> 1.00 1.00 1.01 >> Phase 2 Subphase 1 Iteration 100 Progress: 7% >>
> theta 1.9283 -0.0182 0.5386 >> ac 1.00 1.00 1.01 >> Phase 2 Subphase
> 1 Iteration 110 Progress: 7% >> theta 2.019 -0.019 0.585 >> ac 1.00
> 1.00 1.01 >> Error in x$FRAN(zsmall, xsmall) :
> >> Unlikely to terminate this epoch: more than 1000000 steps >>
> Calls: siena07 ... proc2subphase -> doIterations -> <Anonymous> ->
> .Call >> Execution halted >> >> >> If you got to the end of this
> post, I appreciate your attention and any advice you might have!
> >> Best,
> >> Jennifer Victor
> >> ___________________________________________
> >> Jennifer Nicoll Victor
> >> Assistant Professor
> >> Department of Political Science
> >> University of Pittsburgh
> >> 4600 Wesley W. Posvar Hall
> >> (412) 624-7204
> >> E-mail: jnvictor at pitt.edu
> >> Homepage:
> http://www.polisci.pitt.edu/person/jennifer-nicoll-victor
> >>
> >>
> >> -----Original Message-----
> >> From: rsiena-help-bounces at r-forge.wu-wien.ac.at
> [mailto:rsiena-help-bounces at r-forge.wu-wien.ac.at] On Behalf Of Ruth
> Ripley >> Sent: Friday, March 09, 2012 8:39 AM >> To:
> rsiena-help at r-forge.wu-wien.ac.at >> Subject: Re: [Rsiena-help] What
> does this error mean?
> >>
> >> Dear Jennifer,
> >>
> >> The errors both indicate that you are fitting a model which is too
> >> complicated for your data. I suggest you start with just the
> default >> effects and add others one by one to see which one is
> giving the problem.
> >>
> >> Regards,
> >>
> >> Ruth
> >>
> >> On 09/03/2012 02:45, Victor, Jennifer Nicoll wrote:
> >>> After letting my RSiena model for 12 hours on a high memory
> machine I received the following output. What advice can you provide
> to help me overcome this error?
> >>>
> >>>> myusmodel<-sienaModelCreate(useStdInits=FALSE,
> projname='siena_us_01') >>>> usans01<-siena07(myusmodel,
> data=myusdata,effects=myeffus) >>> No X11 device available, forcing
> use of batch mode >>> Start phase 0 >>> theta: 0.0342 0.0000 0.0000
> 0.0000 0.0000 0.0000 0.0000 0.0000
> 0.0000 0.0000 0.0000
> >>>
> >>> Start phase 1
> >>> Phase 1 Iteration 1 Progress: 0%
> >>> Phase 1 Iteration 2 Progress: 0%
> >>> Phase 1 Iteration 3 Progress: 0%
> >>> Phase 1 Iteration 4 Progress: 0%
> >>> Phase 1 Iteration 5 Progress: 0%
> >>> Phase 1 Iteration 10 Progress: 0% >>> Phase 1 Iteration 15
> Progress: 0% >>> Phase 1 Iteration 20 Progress: 1% >>> Phase 1
> Iteration 25 Progress: 1% >>> Phase 1 Iteration 30 Progress: 1% >>>
> Phase 1 Iteration 35 Progress: 1% >>> Phase 1 Iteration 40 Progress:
> 1% >>> Error in solve.default(z$dfra) :
> >>> system is computationally singular: reciprocal condition number =
> 4.58224e-20
> >>> theta: 0.03416 0.01123 -0.01203 -0.00364 -0.17691 -0.40977
> 0.23908
> -0.14365 0.25761 1.00000 0.21070
> >>>
> >>> Start phase 2.1
> >>> Phase 2 Subphase 1 Iteration 1 Progress: 14% >>> Phase 2
> Subphase 1 Iteration 2 Progress: 14% >>> theta 0.04034 0.01130
> -0.01206 -0.00342 -0.17012 -0.40569 0.24425
> -0.13712 0.25683 0.99919 0.20914
> >>> ac 1.063 1.049 1.006 1.044 1.025 1.063 1.067 1.065 0.651 0.954
> 0.866 >>> Phase 2 Subphase 1 Iteration 3 Progress: 14% >>> Phase 2
> Subphase 1 Iteration 4 Progress: 14% >>> theta 0.05223 0.01144
> -0.01213 -0.00299 -0.15670 -0.39786 0.25405
> -0.12452 0.25377 0.99722 0.20489
> >>> ac 1.136 1.112 1.013 1.096 1.105 1.134 1.161 1.137 0.690 0.984
> 0.881 >>> Phase 2 Subphase 1 Iteration 5 Progress: 14% >>> Phase 2
> Subphase 1 Iteration 6 Progress: 14% >>> theta 0.06332 0.01158
> -0.01221 -0.00256 -0.14449 -0.39056 0.26230
> -0.11261 0.24601 0.99419 0.19779
> >>> ac 1.137 1.140 0.936 1.145 1.104 1.136 1.164 1.140 0.863 0.962
> 0.967 >>> Phase 2 Subphase 1 Iteration 7 Progress: 14% >>> Error in
> x$FRAN(zsmall, xsmall) :
> >>> Unlikely to terminate this epoch: more than 1000000 steps >>>
> Calls: siena07 ... proc2subphase -> doIterations -> <Anonymous> ->
> .Call >>> Execution halted >>> >>> >>> >>> Jennifer N. Victor
> >>> Assistant Professor of Political Science >>> University of
> Pittsburgh >>> 4600 Posvar Hall >>> jnvictor at pitt.edu >>> (412)
> 624-7204 >>> _______________________________________________
> >>> Rsiena-help mailing list
> >>> Rsiena-help at lists.r-forge.r-project.org
> >>>
> https://lists.r-forge.r-project.org/cgi-bin/mailman/listinfo/rsiena-he
> lp
> >>
> >>
> >> _______________________________________________
> >> Rsiena-help mailing list
> >> Rsiena-help at lists.r-forge.r-project.org
> >>
> https://lists.r-forge.r-project.org/cgi-bin/mailman/listinfo/rsiena-he
> lp
> >
> >
> >
--
Ruth M. Ripley, Email:ruth at stats.ox.ac.uk
Dept. of Statistics, http://www.stats.ox.ac.uk/~ruth/
University of Oxford, Tel: 01865 282857
1 South Parks Road, Oxford OX1 3TG, UK Fax: 01865 272595
_______________________________________________
Rsiena-help mailing list
Rsiena-help at lists.r-forge.r-project.org
https://lists.r-forge.r-project.org/cgi-bin/mailman/listinfo/rsiena-help
Nuffield College is a Registered Charity No. 1137506. Registered Office: Nuffield College, New Road, Oxford, OX1 1NF
Nuffield College is a Registered Charity No. 1137506. Registered Office: Nuffield College, New Road, Oxford, OX1 1NF
More information about the Rsiena-help
mailing list