PMC:2846953 / 15503-18071
Annnotations
{"target":"https://pubannotation.org/docs/sourcedb/PMC/sourceid/2846953","sourcedb":"PMC","sourceid":"2846953","source_url":"https://www.ncbi.nlm.nih.gov/pmc/2846953","text":"Variable names and naming conventions for the final models\nTable 2 presents descriptive names and abbreviated names for all untransformed variables that appear in at least one of the final models. Units (for continuous variable) and category names (for categorical variables), are also provided.\nVariable names in final regression models will be of one of the following forms\n1. \u003cv.name\u003e\n2. \u003cv.name\u003e.EQ.xxx, \u003cv.name\u003e.GT.xxx, \u003cv.name\u003e.LT.xxx, \u003cv.name\u003e.GE.xxx, \u003cv.name\u003e.LE.xxx\n3. \u003cv.name\u003e.IFLT.xxx, \u003cv.name\u003e.IFGT.xxx\nHere, \u003cv.name\u003e is the short name of a variable from Table 2 and xxx is a number or category selected by MARS. Form 1 is used for variables that enter into the MARS models as linear functions. Form 2 is used to describe dummy variables formed from a categorical variable. In this context \"EQ\", \"GT\", \"LT\", \"GE\", \"LE\" stand for \"equal\", \"greater than\", \"less than\", \"greater than or equal to\", and \"less than or equal to\". For example, Race.EQ.BLACK designates a variable that is one for blacks and zero otherwise. Alcohol consumption.GT.1 PER WEEK designates a variable that is one for subjects who have more than one drink per week and 0 otherwise. Form 3 is used for transformed continuous variables. For example, CRCL.IFGT.100 (read CRCL if greater than 100) indicates a variable that is equal to CRCL if CRCL is greater than 100 and 100 otherwise. Similarly, CRCL.IFLT.100 is a variable that is equal to CRCL when CRCL is \u003c 100 and equal to 100 otherwise.\nUsing this notation, regression coefficients can be interpreted as slopes. Positive slopes indicate that the dependent variable increases as the predictor variable increases and negative slopes indicate that the dependant variable decreases as the predictor variable increases. Variables of Form 3 are called transformations of the original variable and are thought of as the result of applying a function (or transformation) to the original variable. Other commonly used transformations that do not appear in our regression output are square root, power, and log transformations.\nCollectively, interactions and variables of forms 1, 2, and 3 are called basis functions. The MARS procedure, to be discussed below, may be thought of as a procedure for selecting an optimal set of basis functions that are to be used as the predictor of a given dependent variable. The basis functions used by MARS differ slightly from the basis functions described here. We have chosen to use the basis functions as described here in preference to those used by MARS, because the former seem easier to interpret.","divisions":[{"label":"title","span":{"begin":0,"end":58}},{"label":"p","span":{"begin":59,"end":295}},{"label":"p","span":{"begin":296,"end":375}},{"label":"p","span":{"begin":376,"end":387}},{"label":"p","span":{"begin":388,"end":474}},{"label":"p","span":{"begin":475,"end":514}},{"label":"p","span":{"begin":515,"end":1473}},{"label":"p","span":{"begin":1474,"end":2054}}],"tracks":[]}