# Standardizing coefficients of Logistic Regression Model

#### Kiara_Q

##### New Member
While I'm trying to interpret and use coefficients of Logistic Regression Model, there are two set of problems that I'm facing
Standardizing and bringing to a common scale, all the coefficients? Handling both negative and positive coefficients even after standardization? To explain with an example say with independent variable - var1, var2, var3 model has generated coefficient value of a,-b,c respectively So, to solve 1st part of question, I have scaled them by using partial standardization, agresti coefficients method and multiplied each coefficent value with standard deviation of the sample of that variable So say std1,std2,std3 are standard deviation of the three samples of var1,var2,var3 my transformed coefficients comes out to be -astd1 for var1,-bstd2 for var2, c*std3 for var3.
But now coming to part 2 of problem even after standardizing I have one coefficient that has negative value. I want to scale them in such a way that all the coefficients gets transformed to a positive value and at same time have a common scale so that they can be processed and compared together. Can you please help if you know of possible way of handling this?

#### noetsi

##### Fortran must die
Why do you only want coefficients to be positive? I have never seen anyone do that before. If any of you coefficients are dummy variables then standardization is doubtful according to some experts, although that is a different issue.

#### obh

##### Active Member
In the logistic regression, The sign of the coefficient is related to the effect of the predictor on the odds of the event, related to the base event.
Positive coefficient - say increasing the predictor will increase the odds of the event related to the base event
Negative coefficient - say increasing the predictor will increase the odds of the event related to the base event

#### Kiara_Q

##### New Member
Why do you only want coefficients to be positive? I have never seen anyone do that before. If any of you coefficients are dummy variables then standardization is doubtful according to some experts, although that is a different issue.

In the logistic regression, The sign of the coefficient is related to the effect of the predictor on the odds of the event, related to the base event.
Positive coefficient - say increasing the predictor will increase the odds of the event related to the base event
Negative coefficient - say increasing the predictor will increase the odds of the event related to the base event
I understand that the coefficients sign indicates their positive and negative correlation with the output. The reason why I'm looking for a positive transformation is specific to the problem I'm solving
Independent variables set that I'm using in my model are of nature such that they are present for two sets of channel say channel 1 and second channel 2. By that I mean if var1, var2,var3 as highlighted above are input independent variables in my model, then total # of entries in the model will be 6: var1_chan1, var1_chan2, var2_chan1, var2_chan2,var3_chan1,var3_chan2. Now corresponding to each variable we will have coefficients, so a total of 6 coefficients. To take into account the effect of each channel, when I add up the coefficients for each channel, sometimes the magnitude of -ve coefficients overpower +ve for a channel and in such case comparison of the coefficients contribution across channel i.e for channel 1, channel 2 becomes difficult.

Foe ex: if for channel 1 adding the standardised coefficients of variables : coeff(var1_chan1)-coeff(var2_chan1)+coeff(var3_chan1) = -ve
for channel 2 adding the standardised coefficients of variables : coeff(var1_chan2)-coeff(var2_chan2)+coeff(var3_chan2) = +ve

In such a case how do we compare +ve and -ve value for both the channels. One way is to take contribution of -ve coefficients as zero which would undermine the impact even though in opposite way that var2 is having.

I hope I was able to convey the idea behind transformation of such variable. Can you suggest how can we attribute and split the contributions across channel in such a scenario

#### hlsmith

##### Less is more. Stay pure. Stay poor.
It is not easy to following the context and purposes related to your inquires. I would recommend just writing out the project content. What are you trying to do? When you started writing about the channels I questioned the utility of logistic reg for you or if they represented mediators. It was unclear.

Just tell us about what you are doing and don't try to mask the content. I have found when people try to rephrase context that a lot can get lost. We will either know what you are writing about or not.