Ggtcf

What is the tangent at a sharp point on a curve?

Extract substring according to regexp with sed or grep

Can a Knock spell open the door to Mordenkainen's Magnificent Mansion?

How can I, as DM, avoid the Conga Line of Death occurring when implementing some form of flanking rule?

Should a narrator ever describe things based on a character's view instead of facts?

New Order #2: Turn My Way

Is this saw blade faulty?

Not hide and seek

Would this string work as string?

Is there a distance limit for minecart tracks?

How would a solely written language work mechanically

How to get directions in deep space?

Capacitor electron flow

Why didn't Voldemort know what Grindelwald looked like?

Derivative of an interpolated function

Relations between homogeneous polynomials

Put the phone down / Put down the phone

What can I do if I am asked to learn different programming languages very frequently?

Do I have to take mana from my deck or hand when tapping this card?

Air travel with refrigerated insulin

Highest stage count that are used one right after the other?

How do I lift the insulation blower into the attic?

Pre-Employment Background Check With Consent For Future Checks

Why is indicated airspeed rather than ground speed used during the takeoff roll?

Backtransforming response for plotting allEffects

backtransform `scale()` for plottingPlot two graphs in same plot in RHow to set limits for axes in ggplot2 R plots?Save plot to image file instead of displaying it using Matplotlibvisreg plot binomial distributionCannot run glmer models with na.action=na.fail, necessary for MuMIn dredge functionMeaning of SE for lsmeans type=“response”ggplot: rescale axis (log) and cut axisCorrect way to scale for multilevel regression using lmer [R]scale on effect plot using GLMScaling predictors in lme4 glmer doesn't resolve eigenvalue warnings; neither does alternative optimization

1

I have fitted a few models using glmer from thelme4 package and did model selection with MuMIn::dredge() followed by model averaging. My response is frequency of occurrence of an animal in front of camera traps (continuous, number of photos per 100 hours of camera time); the predictors are minimum distances from a camera trap to landscape categories in meters, sometimes with an interaction with the day phase (day/night).

The response shows Gamma-distribution, but to make the model work I had to transform my response by adding 1, as the Gamma distribution can't handle non-positive data and quite a few of my data points were zeros as shown in this histogram:

Also, I had to scale the predictors, otherwise I got convergence warnings like

Model is nearly unidentifiable: large eigenvalue ratio - Rescale variables?

I did this with scale(..., center = TRUE).

My model formula looks like this:

mod <- glmer(Species_Frequency ~ scaled_Minimal_Distance1 + 
 scaled_Minimal_Distance2 + 
 (scaled_Minimal_Distance3 * Dayphase), 
data = data, 
na.action = na.fail, 
family = Gamma(link = "log"), 
control = glmerControl(optimizer = "bobyqa"))

Now I want to plot the results of mod using plot(allEffects(mod)).

For plotting, I am backscaling my predictors in the allEffects-object by accessing

allEffects(mod)[["scaled_Minimal_Distance1"]][["x"]][["scaled_Minimal_Distance1"]] and doing as shown Here. I would also like to back-transform my response, to make the plot more understandable. Otherwise one would always have to keep in mind that 1 on the y-axis is actually 0, 2 is 1 etc. Can someone help me on how to do this, as I hit a roadblock and can't seem to figure it out myself.

Thank you very much in advance.

r plot

share|improve this question

edited Mar 8 at 20:43

Ben Bolker

135k13228320

asked Mar 7 at 20:31

MatthiasMatthias

63

• 
  
  
  

  
  1
  

  
  
  
  
  
  

  
  
  What do you mean by "frequency of occurrence"? You seem to say that the responses are 0,1,2,... which you transform to 1,2,3,... This seems more like number of occurrences (and counts are discrete while the Gamma distribution is continuous).
  
  – dipetkov
  Mar 7 at 20:57
  

  
  
  
  


• 
  
  
  

  
  

  
  
  
  
  
  

  
  
  What is the ratio of zero to non-zero? It might be worth looking at a Poisson or Negative Binomial distribution, they have ways of dealing with zeroes - en.wikipedia.org/wiki/Zero-inflated_model
  
  – hrabel
  Mar 7 at 21:16
  

  
  
  
  


• 
  
  
  

  
  

  
  
  
  
  
  

  
  
  Please see my edits. I have misformulated this, the "1 is 0 etc" thing described the values on the y-axis, my frequency is continuous. @hrabel I have added a histogram of my non-transformed data. I tried poisson, but as it is non-integer, it didn't work. I should maybe add, that I am quite unexperienced in statistics.
  
  – Matthias
  Mar 8 at 7:35
  
  

  
  
  
  

add a comment | 

1

I have fitted a few models using glmer from thelme4 package and did model selection with MuMIn::dredge() followed by model averaging. My response is frequency of occurrence of an animal in front of camera traps (continuous, number of photos per 100 hours of camera time); the predictors are minimum distances from a camera trap to landscape categories in meters, sometimes with an interaction with the day phase (day/night).

The response shows Gamma-distribution, but to make the model work I had to transform my response by adding 1, as the Gamma distribution can't handle non-positive data and quite a few of my data points were zeros as shown in this histogram:

Also, I had to scale the predictors, otherwise I got convergence warnings like

Model is nearly unidentifiable: large eigenvalue ratio - Rescale variables?

I did this with scale(..., center = TRUE).

My model formula looks like this:

mod <- glmer(Species_Frequency ~ scaled_Minimal_Distance1 + 
 scaled_Minimal_Distance2 + 
 (scaled_Minimal_Distance3 * Dayphase), 
data = data, 
na.action = na.fail, 
family = Gamma(link = "log"), 
control = glmerControl(optimizer = "bobyqa"))

Now I want to plot the results of mod using plot(allEffects(mod)).

For plotting, I am backscaling my predictors in the allEffects-object by accessing

allEffects(mod)[["scaled_Minimal_Distance1"]][["x"]][["scaled_Minimal_Distance1"]] and doing as shown Here. I would also like to back-transform my response, to make the plot more understandable. Otherwise one would always have to keep in mind that 1 on the y-axis is actually 0, 2 is 1 etc. Can someone help me on how to do this, as I hit a roadblock and can't seem to figure it out myself.

Thank you very much in advance.

r plot

share|improve this question

edited Mar 8 at 20:43

Ben Bolker

135k13228320

asked Mar 7 at 20:31

MatthiasMatthias

63

• 
  
  
  

  
  1
  

  
  
  
  
  
  

  
  
  What do you mean by "frequency of occurrence"? You seem to say that the responses are 0,1,2,... which you transform to 1,2,3,... This seems more like number of occurrences (and counts are discrete while the Gamma distribution is continuous).
  
  – dipetkov
  Mar 7 at 20:57
  

  
  
  
  


• 
  
  
  

  
  

  
  
  
  
  
  

  
  
  What is the ratio of zero to non-zero? It might be worth looking at a Poisson or Negative Binomial distribution, they have ways of dealing with zeroes - en.wikipedia.org/wiki/Zero-inflated_model
  
  – hrabel
  Mar 7 at 21:16
  

  
  
  
  


• 
  
  
  

  
  

  
  
  
  
  
  

  
  
  Please see my edits. I have misformulated this, the "1 is 0 etc" thing described the values on the y-axis, my frequency is continuous. @hrabel I have added a histogram of my non-transformed data. I tried poisson, but as it is non-integer, it didn't work. I should maybe add, that I am quite unexperienced in statistics.
  
  – Matthias
  Mar 8 at 7:35
  
  

  
  
  
  

add a comment | 

I have fitted a few models using glmer from thelme4 package and did model selection with MuMIn::dredge() followed by model averaging. My response is frequency of occurrence of an animal in front of camera traps (continuous, number of photos per 100 hours of camera time); the predictors are minimum distances from a camera trap to landscape categories in meters, sometimes with an interaction with the day phase (day/night).

The response shows Gamma-distribution, but to make the model work I had to transform my response by adding 1, as the Gamma distribution can't handle non-positive data and quite a few of my data points were zeros as shown in this histogram:

Also, I had to scale the predictors, otherwise I got convergence warnings like

Model is nearly unidentifiable: large eigenvalue ratio - Rescale variables?

I did this with scale(..., center = TRUE).

My model formula looks like this:

mod <- glmer(Species_Frequency ~ scaled_Minimal_Distance1 + 
 scaled_Minimal_Distance2 + 
 (scaled_Minimal_Distance3 * Dayphase), 
data = data, 
na.action = na.fail, 
family = Gamma(link = "log"), 
control = glmerControl(optimizer = "bobyqa"))

Now I want to plot the results of mod using plot(allEffects(mod)).

For plotting, I am backscaling my predictors in the allEffects-object by accessing

allEffects(mod)[["scaled_Minimal_Distance1"]][["x"]][["scaled_Minimal_Distance1"]] and doing as shown Here. I would also like to back-transform my response, to make the plot more understandable. Otherwise one would always have to keep in mind that 1 on the y-axis is actually 0, 2 is 1 etc. Can someone help me on how to do this, as I hit a roadblock and can't seem to figure it out myself.

Thank you very much in advance.

r plot

share|improve this question

edited Mar 8 at 20:43

Ben Bolker

135k13228320

asked Mar 7 at 20:31

MatthiasMatthias

63

mod <- glmer(Species_Frequency ~ scaled_Minimal_Distance1 + 
 scaled_Minimal_Distance2 + 
 (scaled_Minimal_Distance3 * Dayphase), 
data = data, 
na.action = na.fail, 
family = Gamma(link = "log"), 
control = glmerControl(optimizer = "bobyqa"))

share|improve this question

edited Mar 8 at 20:43

Ben Bolker

135k13228320

asked Mar 7 at 20:31

MatthiasMatthias

63

share|improve this question

edited Mar 8 at 20:43

Ben Bolker

135k13228320

asked Mar 7 at 20:31

MatthiasMatthias

63

edited Mar 8 at 20:43

Ben Bolker

135k13228320

edited Mar 8 at 20:43

Ben Bolker

135k13228320

asked Mar 7 at 20:31

MatthiasMatthias

63

asked Mar 7 at 20:31

MatthiasMatthias

63