[100 Multilevel Questions]How do I incorporate dummy's in R?

是 否 +2 论坛币

k人 参与回答

经管之家送您一份

应届毕业生专属福利!

求职就业群

赵安豆老师微信：zhaoandou666

经管之家联合CDA

送您一个全额奖学金名额~ !

立即领取

感谢您参与论坛问题回答

经管之家送您两个论坛币！

+2 论坛币

I made several dummy variables in SPSS and was wondering if I could use the same ones in R? Is there any special code you need to put before the variables (like factor(...) for categorical) or can you just add them to your model without modifying?I need to use R because I'm doing a multilevel analysis and I want to use both individual as contextual variables.Thanks in advance for your help!

扫码加我 拉你入群

请注明：姓名-公司-职位

以便审核进群资格，未注明则拒绝

分享0 收藏0 回帖

关键词：Incorporate Multilevel questions Corporate question individual modifying because several special

To make R recognize that your variable is categorical and let it create dummies when building a model, the variable should be of class factor. You can check the class by class(object), or by str(df) if its a data frame. If it is not a factor but a numeric variable you can convert it by the factor() command. Show your code if things are not working out still.

You have several options here. First, as suggested by @Edwin, you could use the factor() function to convert a categorical variable to the factor class. R will automatically dummify factors in most linear models (this can be seen by inspecting the model object).

Of course, I don't think anything precludes you from simply including the 1/0 dummies in linear models, as they are allowed per definition. Of course, then you need to manually keep track of them, in relation to their meaning and reference category.

Therefore, the first strategy is probably the most convenient.

To somewhat expand the answer, there are several dummification options in R. The default is dummy coding achieved with contr.treatment(). This will use 1/0 dummies, and it is the default option for unordered factors. See the respective options() output part to verify. There are nevertheless other types of contrasts that can be of use. Deviation coding, in my experience, is often a good candidate. It differs from the default manner of contrast setting in that instead of a single reference category a group mean is taken for pairwise comparisons. This may be interesting in cases like comparing between countries (or states), when there is no intrinsic reason to compare everything to one specific country, but comparison to the mean of all countries makes sense. This type of contrast is set by contr.sum(). Alternatively, you can construct your own custom matrix by hand and supply it by contrasts(x) <- MyMatrix.

More information on contrast types in R can be found here.