Is there a way to force nleqslv to find positive solutions?

Question

I'm trying to solve a non-linear equation with nleqslv, and I know a priori if I want a positive or negative solution (it's a dataset on choice under risk, and I'm trying to compute the risk aversion coefficient for each individual under CRRA assumption. Since I can observe the DMs' choices, I already know if each DM is risk averse or not). Is there any way to enforce it?

I know I could try with different initial values; however, I'd like to find another way as I'm using nleqslv in a for loop (I must compute one solution for each observation) and I can't find an initial guess that works for everyone.

My code is the following:

`bernoulli <- function(x, r) {   
  ifelse(x>=0,((x+1)^(1-r)-1)/(1-r), -((-x+1)^(1-r)-1)/(1-r) )     
}

bernoulli.log <- function(x) {  
  ifelse(x>=0, log(x+1), -log(-x+1) )}  


mydata$alpha_crra <- NA  


for (i in 1:nrow(mydata))  {  

  indiff.eq.crra <- function(r)  {  

  return(bernoulli(mydata$CE[i], r) - mydata$p[i]*bernoulli(mydata$win[i], r)   
            - (1-mydata$p[i])*bernoulli(mydata$lose[i], r))  

}  

  mydata$alpha_crra[i] <- ifelse(mydata$riskneutral[i] == 1, 0,  
                                 ifelse(abs(bernoulli.log(mydata$CE[i]) -   
                                              mydata$p[i]*bernoulli.log(mydata$win[i]) -  
                                              (1-mydata$p[i])*bernoulli.log(mydata$lose[i])) < 0.001 &  
                                          mydata$riskaverse[i] == 1, 1,  
                                        nleqslv(-5, indiff.eq.crra)$x))  


}` 

where:

mydata$win[i]= the high payoff of the lottery (can change depending on the observation) mydata$lose[i]= the low payoff of the lottery mydata$p[i] = probability to win the high payoff mydata$CE[i]= the Certainty Equivalent stated by DM i mydata$riskneutral[i] = a dummy variable = 1 if i is risk neutral, 0 otherwise mydata$riskaverse[i] = a dummy variable = 1 if i is risk averse, 0 otherwise