![nonmem model for categorical outcome nmuser nonmem model for categorical outcome nmuser](https://monolix.lixoft.com/wp-content/uploads/sites/17/2018/03/shrinkage_table.png)
This is accomplished by transforming the raw outcome values into probability (for one of the two categories), odds or odds ratio, and log odds (literally the ‘log’ of the odds / odds ratio). Logistic regression is used to regress categorical and numeric variables onto a binary outcome variable. In this article the term logistic regression (Cox, 1958) will be used for binary logistic regression rather than also including multinomial logistic regression.
![nonmem model for categorical outcome nmuser nonmem model for categorical outcome nmuser](https://www.researchgate.net/profile/Itziar-Oteo/publication/225097865/figure/tbl2/AS:667828042612742@1536234021379/NONMEM-parameter-estimates-from-the-base-model_Q640.jpg)
The t-tests are simply testing whether the coefficients are different than zero (here, all four are). So the coefficients for this model are the average outcome variable value for the category (or level) of the predictor variable. To interpret all four coefficients (listed in the 'no-intercept' model).we would say that all cases with a value of "d" on 'x2' would be predicted to have a value of 4.5 on 'y' because, that is the average of the “d” cases on ‘y’. To see all the coefficients, we can run a no-intercept model or simply plot the two variables. The t-test is simply testing if the difference between, say the category 'a' coefficient and the reference category, 'd' is different than zero: 4.50 - 1.50 = -3.00 is that absolute difference greater than zero? Yes, p = 0.0000159. Likewise, the mean of "b" is 2.0 units less than the mean of "d" and the mean of "c" is 1.0 units less than the mean of "d" - because "d" is the reference category in the linear regression and the negative coefficients represent the less than in the interpretation. So, looking at the 'x2' model, directly above we see that the mean (y-value) of category, or level, "a" is 3.0 units less than the mean (y-value) of "d" (which is listed as the intercept). Residual standard error: 0.4 on 8 degrees of freedom The reference category, which was not user-specified, is “a” because it is alphabetically first of the levels. It has four levels: “a”, “b”, “c”, and “d”. Consider the factor ‘x1’ below, which is created by replicating the first four letters of the alphabet three times. By default R uses the alpha-numerically first category as the reference category (e.g. In other words, the other categories are compared to the reference. A ‘reference’ category is so named and identified as a category of comparison for the other categories. The categories of a factor are identified as ‘levels’ of the factor. strictly discrete categorical variables).
![nonmem model for categorical outcome nmuser nonmem model for categorical outcome nmuser](https://www.frontiersin.org/files/Articles/673492/fphar-12-673492-HTML/image_m/fphar-12-673492-g002.jpg)
Throughout this article we will be dealing with unordered factors (i.e. The R language identifies categorical variables as ‘factors’ which can be ‘ordered’ or not. First, we must understand how R identifies categorical variables. The primary purpose of this article is to illustrate the interpretation of categorical variables as predictors and outcome in the context of traditional regression and logistic regression. 1, 2018, Research Matters, Benchmarks Online By Jonathan Starkweather, Ph.D., consultant, Data Science and Analytics | Nov.