CODEWITHSUNDEEP
r
user2946746
user2946746

Reputation: 1780

Fitting curves with DRC package in R?

I'm trying to fit curves with the DRC package in R.

Example: 

 x_yrs<-c(2005, 2006, 2007, 2008, 2009, 2010, 2011, 2012, 2013, 2014, 
     2015, 2016, 2017)
 y<-c(1.89, 0.34, 0.47, 2.46, 2.13, 7.49, 47.24, 117.84, 202.8, 322.7, 
 540.72, 744.22, 1148.7)

 MaxPop<-110000
 Y_Adj<-y/MaxPop

 EV<-drm(y~ x_yrs,fct = LL.3(fixed = c(NA, NA, NA)))
 plot(EV, broken = TRUE, type = "all")

 EV<-drm(y~ x_yrs,fct = LL.5(fixed = c(NA, NA, NA, NA, NA)))
 plot(EV, broken = TRUE, type = "all")

 x_yrs_Adj<- x_yrs-2004
 EV<-drm(Y_Adj~ x_yrs_Adj,fct = LL.5(fixed = c(NA, NA, NA, NA, NA)))
 plot(EV, broken = TRUE, type = "all",xlim = c(0, 40), ylim = c(0, 1))

I would like to max value of the curve to be "1" or the "MaxPop" ie as the upper asymptote.

How would I go about changing the drm model to accomplish this?

Upvotes: 0

Views: 938

Answers (1)

Maurits Evers
Maurits Evers

Reputation: 50668

"I would like to set the future population size to reach 110,000." I don't think it will be possible to fit a model with that constraint based on the data you give. The response that you have for the support of the function doesn't even get near to that (potentially?) asymptotic region. So I think you need to rethink your approach.

That aside, in drc you can realise constraints by specifying values for specific parameters through the fixed function argument.

EV <- drm(Y_Adj ~ x_yrs_Adj, fct = LL.5(fixed = c(NA, 0, 1, NA, NA)))

You can find out about the individual parameters if you do e.g. ?LL.5:

  LL.5(fixed = c(NA, NA, NA, NA, NA), names = c("b", "c", "d", "e", "f"), ...)

[...]

The five-parameter logistic function is given by the expression

      f(x) = c + \frac{d-c}{(1+\exp(b(\log(x)-\log(e))))^f}

So in this case, we set c to zero and then fix d = 1.

Let's show the plot

plot(EV, broken = TRUE, type = "all", xlim = c(0, 40000), ylim = c(0, 1))

You can see the issue here. As you don't have any support of values x_yrs_Adj closer to the function's asymptotic behaviour, your fit (and the resulting estimated parameters) will be poor.

enter image description here

Upvotes: 1

Related Questions