Re: [structure-group] L(K) calculation

[Marco Pessoa]

LnP(D) would be the first item on your list: "Estimated Ln Prob of Data", with value -220551.5


Marco Pessoa
Pesquisador A - Biologia Molecular
Embrapa Cerrados
BR 020-Km 18, Brasília, DF - Brasil
CEP 73310-970
Fone: +55-61-3388-9889
http://www.cpac.embrapa.br

Em 22/06/2012, às 16:18, Bao Lam Huynh escreveu:

> To whom It May Concern:
>
> Hi,
>
> I am calculating delta K to decide the optimum K value fro Structure following the instruction in EVANNO et al. 2005 Molecular Ecology 14: 2611–2620. However, this paragraph makes me confused:
>
> "The model choice criterion implemented in structure to detect the true K is an estimate of the posterior probability of the data for a given K, Pr(X|K) (Pritchard et al 2000). This value, called ‘LnP(D)’ in structure output, is obtained by first computing the log likelihood of the data at each step of the MCMC. Then the average of these values is computed and half their variance is subtracted to the mean. This gives ‘Ln P(D)’, the model choice criterion to which we refer as L(K) afterwards."
>
> For example, below is an output for K=3 from Structure, I am not sure if LnP(D) = L(K) = -220138.4, or = -220138.4 - 826.1 according to the above paragraph?
>
> Estimated Ln Prob of Data = -220551.5
> Mean value of ln likelihood = -220138.4
> Variance of ln likelihood = 826.1
> Mean value of alpha = 0.1939
>
> Basically, I would like to know what is L(K) from this Structure output as it is a component in the deltaK formula. I would very much appreciate if you could help me on this.
>
> Sincerely
> Lam
>
> --
> You received this message because you are subscribed to the Google Groups "structure-software" group.
> To view this discussion on the web visit https://groups.google.com/d/msg/structure-software/-/qCtznfG61mIJ.
> To post to this group, send email to structure...@googlegroups.com.
> To unsubscribe from this group, send email to structure-softw...@googlegroups.com.
> For more options, visit this group at http://groups.google.com/group/structure-software?hl=en.

[ywee chieh tay]

Hi Lam,

You can also use structure harvester to do the calculations for you: taylor0.biology.ucla.edu/structureHarvester/

It'll save alot of effort especially if you have many runs! ;p

Cheers,
yc

On Sat, Jun 23, 2012 at 3:18 AM, Bao Lam Huynh <baolam...@ucr.edu> wrote:

To whom It May Concern:

Hi,

I am calculating delta K to decide the optimum K value fro Structure following the instruction in EVANNO et al. 2005 Molecular Ecology 14: 2611–2620. However, this paragraph makes me confused:

"The model choice criterion implemented in structure to detect the true K is an estimate of the posterior probability of the data for a given K, Pr(X|K) (Pritchard et al  2000). This value, called ‘LnP(D)’ in structure output, is obtained by first computing the log likelihood of the data at each step of the MCMC. Then the average of these values is computed and half their variance is subtracted to the mean. This gives ‘Ln P(D)’, the model choice criterion to which we refer as L(K) afterwards."

For example, below is an output for K=3 from Structure, I am not sure if LnP(D) = L(K) = -220138.4, or = -220138.4 - 826.1 according to the above paragraph? 

Estimated Ln Prob of Data   = -220551.5

Mean value of ln likelihood = -220138.4

Variance of ln likelihood   = 826.1

Mean value of alpha         = 0.1939

Basically, I would like to know what is L(K) from this Structure output as it is a component in the deltaK formula. I would very much appreciate if you could help me on this.

Sincerely

Lam

--
You received this message because you are subscribed to the Google Groups "structure-software" group.
To view this discussion on the web visit https://groups.google.com/d/msg/structure-software/-/qCtznfG61mIJ.
To post to this group, send email to structure...@googlegroups.com.
To unsubscribe from this group, send email to structure-softw...@googlegroups.com.
For more options, visit this group at http://groups.google.com/group/structure-software?hl=en.

--
yujie

[Ajay jajodia]

Hi..
Structure harvester mention by Lam is..ultimate tool..

Cheers
ajay

[Dyanna L.]

I agree with the others. Check out Structure Harvester. CorrSieve is also pretty good, but doesn't calculate sd.

Dyanna

[Bao Lam Huynh]

Dear YC Tay and all of you who responded to my message. I have tried structure harvester. It is really useful. Thanks so much for kindly letting me know. I apologize for this delaying reply but I have not logged in Google group for a while. I thought the messages would be sent to my email box. Many thanks again. All the best. Lam

On Saturday, June 23, 2012 12:38:08 AM UTC-7, YC Tay wrote:

Hi Lam,

You can also use structure harvester to do the calculations for you: taylor0.biology.ucla.edu/structureHarvester/

It'll save alot of effort especially if you have many runs! ;p

Cheers,
yc

On Sat, Jun 23, 2012 at 3:18 AM, Bao Lam Huynh <baolam...@ucr.edu> wrote:

To whom It May Concern:

Hi,

I am calculating delta K to decide the optimum K value fro Structure following the instruction in EVANNO et al. 2005 Molecular Ecology 14: 2611–2620. However, this paragraph makes me confused:

"The model choice criterion implemented in structure to detect the true K is an estimate of the posterior probability of the data for a given K, Pr(X|K) (Pritchard et al  2000). This value, called ‘LnP(D)’ in structure output, is obtained by first computing the log likelihood of the data at each step of the MCMC. Then the average of these values is computed and half their variance is subtracted to the mean. This gives ‘Ln P(D)’, the model choice criterion to which we refer as L(K) afterwards."

For example, below is an output for K=3 from Structure, I am not sure if LnP(D) = L(K) = -220138.4, or = -220138.4 - 826.1 according to the above paragraph? 

Estimated Ln Prob of Data   = -220551.5

Mean value of ln likelihood = -220138.4

Variance of ln likelihood   = 826.1

Mean value of alpha         = 0.1939

Basically, I would like to know what is L(K) from this Structure output as it is a component in the deltaK formula. I would very much appreciate if you could help me on this.

Sincerely

Lam

--
You received this message because you are subscribed to the Google Groups "structure-software" group.
To view this discussion on the web visit https://groups.google.com/d/msg/structure-software/-/qCtznfG61mIJ.

To post to this group, send email to structure-software@googlegroups.com.
To unsubscribe from this group, send email to structure-software+unsub...@googlegroups.com.

For more options, visit this group at http://groups.google.com/group/structure-software?hl=en.

--
yujie