[R-sig-ME] Error: In mer_finalize(ans) : false convergence (8)
ONKELINX, Thierry
Thierry.ONKELINX at inbo.be
Fri Jul 26 13:35:16 CEST 2013
Sarah,
Always keep the mailing list in cc while replying.
After having a close look to you design I would suggest to go for one of the models below. If you are interested how the effect of TMT1 and TMT2 change in time go for Mod1. Go for Mod2 is you expect that their effect remains constant over time. Otherwise go for Mod3.
lmerdat$SpuplotID <- with(lmerdat, interaction(Block, TMT1, TMT2))
Mod1 <- lmer(log(Y) ~ TMT1 * TMT2 * Year + (1|Block/SubplotID), data = lmerdat, verbose = TRUE)
Mod2 <- lmer(log(Y) ~ TMT1 * TMT2 + (1|Block/SubplotID) + (1|Year), data = lmerdat, verbose = TRUE)
Mod3 <- lmer(log(Y) ~ TMT1 * TMT2 + (1|Block/SubplotID) + (0 + TMT1 : TMT2|Year), data = lmerdat, verbose = TRUE)
Assuming Year is continuous Mod1 requires 10 parameters to be estimated. Your design yields 24 observations per year. Unless you have at least 4 years of data, the model is too complex for your data. You might want to look at the quote of Fisher in my signature...
Best regards,
ir. Thierry Onkelinx
Instituut voor natuur- en bosonderzoek / Research Institute for Nature and Forest
team Biometrie & Kwaliteitszorg / team Biometrics & Quality Assurance
Kliniekstraat 25
1070 Anderlecht
Belgium
+ 32 2 525 02 51
+ 32 54 43 61 85
Thierry.Onkelinx op inbo.be
www.inbo.be
To call in the statistician after the experiment is done may be no more than asking him to perform a post-mortem examination: he may be able to say what the experiment died of.
~ Sir Ronald Aylmer Fisher
The plural of anecdote is not data.
~ Roger Brinner
The combination of some data and an aching desire for an answer does not ensure that a reasonable answer can be extracted from a given body of data.
~ John Tukey
-----Oorspronkelijk bericht-----
Van: Sarah Dryhurst [mailto:s.dryhurst op gmail.com]
Verzonden: vrijdag 26 juli 2013 12:07
Aan: ONKELINX, Thierry
Onderwerp: Re: [R-sig-ME] Error: In mer_finalize(ans) : false convergence (8)
Dear Thierry,
Thanks so much for you reply, Apologies - I should have explained the experiment better. I have:
6 blocks
2 plots per block: one receiving one level of TMT1, one as a TMT1 control
2 subplots per plot: one receiving TMT2, one as TMT2 control.
The response is a measure of biomass (production) in each subplot I tried removing it as a fixed effect as you suggested, and the model does indeed converge. However I'm interested in the effects of both
TMT1 and TMT2 and how they might interact, so I want to retain TMT1 as both a fixed effect and a random effect nested with block (due to the nature of the design)... Perhaps I could remove year as a fixed effect? Forgive me if I am making a silly mistake here: still learning this stuff!
Thanks again
Sarah
On Fri, Jul 26, 2013 at 9:16 AM, ONKELINX, Thierry <Thierry.ONKELINX op inbo.be> wrote:
> Dear Sarah,
>
> Have TMT1 both in the fixed effects and nested in block as a random effect seems a strange to me. This might be the problem for the convergence.
>
> 1) What is the rationale for include TMT1 both in the fixed as the random effects?
> 2) Do you get convergence with
> Mod1 <- lmer(log(Y) ~ TMT1 * TMT2 * Year + (1|Block) + (1|Year), data
> = lmerdat, verbose = TRUE)
> Mod2 <- lmer(log(Y) ~ TMT2 * Year + (1|Block/TMT1) + (1|Year), data =
> lmerdat, verbose = TRUE)
>
> Andpleaseaddsomewhitespacetoyourcode
>
> Best regards,
>
> Thierry
>
> ir. Thierry Onkelinx
> Instituut voor natuur- en bosonderzoek / Research Institute for Nature
> and Forest team Biometrie & Kwaliteitszorg / team Biometrics & Quality
> Assurance Kliniekstraat 25
> 1070 Anderlecht
> Belgium
> + 32 2 525 02 51
> + 32 54 43 61 85
> Thierry.Onkelinx op inbo.be
> www.inbo.be
>
> To call in the statistician after the experiment is done may be no more than asking him to perform a post-mortem examination: he may be able to say what the experiment died of.
> ~ Sir Ronald Aylmer Fisher
>
> The plural of anecdote is not data.
> ~ Roger Brinner
>
> The combination of some data and an aching desire for an answer does not ensure that a reasonable answer can be extracted from a given body of data.
> ~ John Tukey
>
>
> -----Oorspronkelijk bericht-----
> Van: r-sig-mixed-models-bounces op r-project.org
> [mailto:r-sig-mixed-models-bounces op r-project.org] Namens Sarah
> Dryhurst
> Verzonden: donderdag 25 juli 2013 19:51
> Aan: r-sig-mixed-models op r-project.org
> Onderwerp: [R-sig-ME] Error: In mer_finalize(ans) : false convergence
> (8)
>
> Hello all,
>
> I am running an mixed effects model on the attached data. I initially
> used the following code
>
> mod1<-
> lmer(log(Y)~TMT1*TMT2*Year+(1|Block/TMT1)+(1|Year),data=lmerdat,verbos
> e=TRUE)
>
> Which gives me the following:
>
> 0: 77.714051: 0.666667 0.471405 0.333333
> 1: 74.110931: 0.842664 1.45578 0.328587
> 2: 72.264680: 0.440773 1.34787 0.215307
> 3: 72.215476: 0.405289 1.27443 0.208252
> 4: 72.206023: 0.430340 1.25642 0.208492
> 5: 72.203289: 0.423498 1.22704 0.214935
> 6: 72.202721: 0.422680 1.22886 0.209101
> 7: 72.202679: 0.424822 1.23459 0.208315
> 8: 72.202679: 0.424822 1.23459 0.208315
> 9: 72.202679: 0.424822 1.23459 0.208315
> Warning message:
> In mer_finalize(ans) : false convergence (8)
>
>
>
> I thought that I might have too many parameters, so I removed the three way interaction thus:
>
>
>
> mod1<-
> lmer(logY)~TMT1+TMT2+Year+TMT1:TMT2+TMT1:Year+TMT2:Year+(1|Block/TMT1)
> +(1|Year),data=lmerdat,verbose=TRUE)
>
> 0: 82.176953: 0.666667 0.471405 0.333333
> 1: 80.060839: 0.932569 1.42347 0.182093
> 2: 79.408200: 0.00000 1.72496 0.180290
> 3: 78.436511: 0.00107482 1.31540 0.182908
> 4: 78.313534: 0.00586792 1.19163 0.169823
> 5: 78.292542: 0.0255058 1.16060 0.176287
> 6: 78.204264: 0.195335 0.918497 0.214946
> 7: 77.959243: 0.166960 1.02200 0.173634
> 8: 77.325791: 0.416075 1.15366 0.271404
> 9: 77.240857: 0.377681 1.16245 0.202179
> 10: 77.237968: 0.381277 1.16539 0.190858
> 11: 77.237719: 0.380920 1.16760 0.187107
> 12: 77.237718: 0.379746 1.17007 0.187519
> 13: 77.237695: 0.381074 1.16994 0.187148
> 14: 77.237695: 0.380772 1.16969 0.187087
> 15: 77.237695: 0.380778 1.16974 0.187092
> 16: 77.237695: 0.380806 1.16978 0.187098
> 17: 77.237695: 0.380906 1.16978 0.187107
> 18: 77.237695: 0.380999 1.16975 0.187128
> 19: 77.237695: 0.380982 1.16965 0.187105
> 20: 77.237695: 0.380961 1.16967 0.187112
> 21: 77.237695: 0.380961 1.16967 0.187112
> Warning message:
> In mer_finalize(ans) : false convergence (8)
>>
>
>
>
> So the model seemed to get a bit further but still will not converge... I don't want to remove anything else as my model will then not invetsigate the relationships I want it to and I don't think it is overparameterised without that 3 way interaction...
>
> If i multiply Y by 10 it gets a bit further again but still stalls:
>
>
>
>
>> mod1<-
>> lmer(log(I(10*Y))~TMT1+TMT2+Year+TMT1:TMT2+TMT1:Year+TMT2:Year+(1|Blo
>> c
>> k/TMT1)+(1|Year),data=lmerdat,verbose=TRUE)
> 0: 82.176953: 0.666667 0.471405 0.333333
> 1: 80.059473: 0.932336 1.42350 0.181868
> 2: 79.407034: 0.00000 1.72457 0.181162
> 3: 78.529865: 0.00000 1.30356 0.0708756
> 4: 78.320758: 0.000702149 1.19618 0.193718
> 5: 78.308420: 0.00287500 1.16437 0.161311
> 6: 78.298595: 0.0199600 1.14541 0.174745
> 7: 78.198050: 0.0954940 1.07394 0.224686
> 8: 78.095409: 0.159180 1.01562 0.265286
> 9: 77.628631: 0.238546 1.05911 0.235037
> 10: 77.251998: 0.395868 1.12462 0.185342
> 11: 77.239703: 0.373209 1.15530 0.187029
> 12: 77.238348: 0.378481 1.15960 0.186789
> 13: 77.237962: 0.380487 1.16277 0.186797
> 14: 77.237766: 0.380997 1.16609 0.186947
> 15: 77.237720: 0.380749 1.16757 0.187021
> 16: 77.237698: 0.381015 1.16893 0.187108
> 17: 77.237697: 0.380978 1.16905 0.187112
> 18: 77.237696: 0.380954 1.16929 0.187109
> 19: 77.237696: 0.381002 1.16935 0.187111
> 20: 77.237695: 0.381017 1.16942 0.187070
> 21: 77.237695: 0.380947 1.16958 0.187109
> 22: 77.237695: 0.380947 1.16976 0.187103
> 23: 77.237695: 0.380948 1.16975 0.187105
> 24: 77.237695: 0.380948 1.16975 0.187105
> 25: 77.237695: 0.380948 1.16975 0.187105
> 26: 77.237695: 0.380948 1.16975 0.187105
> Warning message:
> In mer_finalize(ans) : false convergence (8)
>
>
>
> Is there a solution to this? Can I trust the model outputs that I am
> getting in spite of this convergence issue? I can't use the suggested
> resolution of dividing or multiplying explanatory variables by
> 10/100/1000 etc as these variables are categorical....
>
> Sorry to post another question about this - I just can't find an answer that fits my data.
>
> Any thoughts would be much appreciated
>
> Best wishes,
>
> Sarah
>
> --
> NERC PhD Student
> Community Ecology and Global Change
> Department of Biology
> Imperial College, London
> email: sarah.dryhurst08 op imperial.ac.uk
> * * * * * * * * * * * * * D I S C L A I M E R * * * * * * * * * * * *
> * Dit bericht en eventuele bijlagen geven enkel de visie van de schrijver weer en binden het INBO onder geen enkel beding, zolang dit bericht niet bevestigd is door een geldig ondertekend document.
> The views expressed in this message and any annex are purely those of the writer and may not be regarded as stating an official position of INBO, as long as the message is not confirmed by a duly signed document.
--
NERC PhD Student
Community Ecology and Global Change
Department of Biology
Imperial College, London
email: sarah.dryhurst08 op imperial.ac.uk
* * * * * * * * * * * * * D I S C L A I M E R * * * * * * * * * * * * *
Dit bericht en eventuele bijlagen geven enkel de visie van de schrijver weer en binden het INBO onder geen enkel beding, zolang dit bericht niet bevestigd is door een geldig ondertekend document.
The views expressed in this message and any annex are purely those of the writer and may not be regarded as stating an official position of INBO, as long as the message is not confirmed by a duly signed document.
More information about the R-sig-mixed-models
mailing list