# Interpreting multinomial logistic regression output in R

I am trying to perform multinomial logistic regression on my data which is as below(just the header). "category" is my target variable and all other variables are independent variables. category has values like 1,2,3,4,5,6,7,8,9. My main motive is to predict category from independent variables. Here is summary of my data set. I used following command to perform multinomial logistic regression:

> mod=multinom(category~hlen+iplen+ipttl+iptype+tcpsport+tcpdport+tcpsec+tcpack+tcpwindow+tcpchksum+date_time, data=train) I got the following output but i don't know how to interpret it? What should be the starting point or is there other way to do so? (For example I know how to interpret linear and logistic regression output to create regression equations)

Please right click any image and select view to see it clearly. Thank you,

• Please replace the images with the actual text. – Sven Hohenstein Dec 17 '12 at 16:28

## 1 Answer

Multinomial logistic regression works like a series of logistic regressions, each one comparing two levels of your dependant variable. Here, category 1 is the reference category.

For example, consider the case where you only have values where category is 1 or 5. (Recode that to 0 and 1, so that you can perform logistic regression.) The coefficients of line 5 of your output represent the variable coefficients such a logistic regression would yield.

• Thanks mzuba, So is it correct to write logit for category 2 as(considering intercept as "ß0", hlen coefficient as "ß1",iplen coefficient as "ß2" and so on....): logit(p)=ß0+ß1(hlen)+ß3(iplen)+ß4(ipttl)-ß5(iptype)-ß6(tcpsport)-ß7(tcpdport)-ß8(tcpsec)-ß9(tcpack)-ß10(tcpwindow)+ß11(tcpchksum)+ß12(date_time) The values of ß1,ß2,ß3...so on, I can get from above output? – user16603 Dec 3 '12 at 17:44
• log[p(category=2)/p(category=1)]=ß0+ß1(hlen)+ß3(iplen)+ß4(ipttl)-ß5(iptype)-ß6(tcpsport)-ß7(tcpdport)-ß8‌​(tcpsec)-ß9(tcpack)-ß10(tcpwindow)+ß11(tcpchksum)+ß12(date_time) The values of ß1,ß2,ß3...so on, I can get from above output? p=probability – user16603 Dec 3 '12 at 21:25
• Yes. The intercept values is you B1, B2 and so on values – user107121 Mar 1 '16 at 23:55