# Linear Regression Data Transformation [duplicate]

My regression model has pulse rate difference (continuous variable) as the DV and Body Mass Index (BMI, continuous variable) and Level of Exercise (low, moderate, high, categorical variable) as the 2 independent variables. Do I start with scatter plots and look for a linear pattern in the plots first or transform the non-normal data first? If all are non-normal (DV and IVs) do they all need to be transformed? n=110.

• These issues have been extremely well covered in other threads. Start by searching our site on obvious keywords like "transform regression normal". That often turns up useful stuff. In this case you might want to exclude "logistic". – whuber Feb 16 '15 at 21:26
• Where to begin? Without a model, how are you assessing normality? If you're just looking at the response variable (DV) on its own, there's no assumption about that. With multiple variables, it can be difficult to assess linearity without adjusting for the other variables. [If you're using the data to choose a model which you then want to apply inference to, you need to account for the effect of that.] -- broadly speaking you should worry most about getting the description of the mean right, but where possible the form of the model should be based on subject area knowledge. – Glen_b Feb 16 '15 at 21:32
• Conceptual regression model: Pulse1 (continuous)= Body Mass Index (continuous) + ModerateExercise (dummy coded) + HighExercise (dummy coded) + error. Research question: Is there a relationship between BMI, exercise level and pulse1? I need to conduct a linear regression. Just want to know what my fist step should be. Is it to check the distribution of each continuous variable and then transform if necessary. Then create scatter plots and then fit the regression line using the ENTER method in SPSS, as this is exploratory? – user3096214 Feb 18 '15 at 11:48