2
$\begingroup$

I have run a linear regression with the following equation (in r):

lm(formula = logTotal ~ Continent + logArea + Method + Servs)

where Total is $/ha/year (numeric), Area is hectare (numeric), Continent and Method are factors and Servs is numeric. It returns the output:

Call:
lm(formula = logTotal ~ Continent + logArea + Method + Servs)

Residuals:
     Min       1Q   Median       3Q      Max 
-0.99416 -0.26931 -0.00622  0.28885  1.19875 

Coefficients:
                       Estimate Std. Error t value Pr(>|t|)    
(Intercept)            -1.82886    0.71446  -2.560 0.016903 *  
ContinentAsia           3.82452    0.60471   6.325 1.28e-06 ***
ContinentAustralasia    4.52516    0.96517   4.688 8.35e-05 ***
ContinentEurope         2.18022    0.48260   4.518 0.000130 ***
ContinentGlobal         2.44750    0.74092   3.303 0.002881 ** 
ContinentNorth America  2.35244    0.55281   4.255 0.000256 ***
ContinentSouth America  3.67853    0.61454   5.986 2.99e-06 ***
logArea                 0.03643    0.03583   1.017 0.318911    
MethodCVM              -0.18171    0.43296  -0.420 0.678300    
MethodOther hedonic    -1.53284    0.79781  -1.921 0.066165 .  
MethodValue Transfer    0.98101    0.29773   3.295 0.002941 ** 
Servs                   0.10723    0.04273   2.509 0.018948 *  
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Residual standard error: 0.552 on 25 degrees of freedom
  (65 observations deleted due to missingness)
Multiple R-squared:  0.7582,    Adjusted R-squared:  0.6518 
F-statistic: 7.127 on 11 and 25 DF,  p-value: 2.501e-05  

I wish to predict Total based on various inputs, however I'm a bit lost on fully understanding the output. If I wished to predict "Total" on the basis of:

Continent:Global, Area:1 hectare, Method:CVM, Servs:11

is the following equation correct?

exp(Total) = 2.44750 + exp(1*0.03643) - 0.18171 + (11*0.10723)

I have read UCLA's statistics help site's FAQ on log transformed regression. I feel like I've oversimplified it but I just keep reading that link over and over and still not fully understanding. Also read How to interpret logarithmically transformed coefficients in linear regression?.

$\endgroup$
2
$\begingroup$

is the following equation correct?

exp(Total) = 2.44750 + exp(1*0.03643) - 0.18171 + (11*0.10723)

No, but close. Assuming you used natural log:

$\ln{Total} = -1.82886+2.44750+\ln(1)\times0.03643-0.18171+11\times0.10723$

Then, exponent on both sides:

$Total = exp^{-1.82886+2.44750+\ln(1)\times0.03643-0.18171+11\times0.10723}$


If your goal is just to predict, the above simple substitution will work fine.

The $exp^{\beta}$ think you read about online is pertinent when you wish to interpret the coefficient individually.

And here provides some resource on dealing with retransformation bias, which I didn't know until Dimitriy pointed that out.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.