4.3 Guidance. It is our intention that the user be given free rein over the construction of the analysis. This is in keeping with having a professional statistician as the target user. Nevertheless, DINDE is constructed so as to inconspicuously guide the analysis where possible. Two different approaches to this task can be taken. In one, the guidance is strictly independent of the data in hand. A metaphor might be the sort of guidance given over the telephone by an experienced statistician to a less experienced one. The advice-giving statistician has no access at at all to the data, not even the subject-matter context from which the data arose. The less experienced statistician only wants to know how one should perform a regression analysis, say, what to look for, and so on. (Argument for the credibility of this metaphor can be found in Oldford and Peters, 1986a.) In DINDE, we mimic this kind of guidance through the selection and design of classes, in particular the Situation classes. Each class represents a stage or element in the overall analysis. Hence, the choice of classes in DINDE represents our cataloguing of the statistical procedures and concepts we regard to be relevant to different kinds of analyses (BivariateRegression, BivariateLeastSquares, UnivariateTimeSeriesAnalysis, etc.). The organization of the methods that a particular class of object responds to indicates those steps that are generally taken next in the analysis. For example, a BivariateLeastSquares object offers a selection of methods which each produce a useful residual plot. In this way, what appears to be merely convenient for the experienced analyst turns out to be guidance for the less experienced. This data-independent channelling of the analysis is made more explicit by adding Suggestions to each DINDEObject. In this way, some text may be given which organizes the alternatives according to various characteristics of the data and analysis to date. The onus is then on the user to determine how the characteristics of the current data and analysis match those outlined in the Suggestions. Self-explanatory examples would also be helpful in this regard. The second approach is the data-dependent one, whereby suggestions made to the user are based upon having the system investigate, and give some interpretation to, the properties of the data in hand. Bivariate regression analysis in REX is an important example of this approach (see Pregibon and Gale, 1984). In DINDE, we plan to adopt this approach in an inconspicuous and quite local manner. It is local in two senses. First, such data checking operations would be attached to particular DINDE classes and would be available to the user as a method of that node type in the analysis. Second, the method itself would have access to only two sources of information › either that attached to the object, or that to be had by querying the user. In particular, no information regarding the analysis map or history would be used. In this way, no attempt is made to provide the user with a global analysis › all suggestions are local and as such the analyst may justifiably choose to ignore them in light of the rest of the analysis. For instance, a BivariateRegression might have a method called CheckFunctionalForm which would try to determine whether different transformations of the X and Y data might be better suited to a linear regression analysis, and provide evidence for any comments made. (Hopefully, the logarithmic transformation would be suggested in the example.) The user would then be free to follow the suggestions or not. While both approaches have merit, we feel that data-independent channelling is more easily implemented and provides much of the guidance that can be competently given (for further discussion, see Oldford and Peters (1986a)). In any case, it seems to be the natural precursor to the data-dependent form of guidance.(LIST ((PAGE NIL (PAPERSIZE Letter FOLIOINFO (ARABIC ) STARTINGPAGE# 23) (0 0 612 792) ((FOLIO NIL (PARALOOKS (QUAD CENTERED) CHARLOOKS (SUPERSCRIPT 0 INVISIBLE OFF SELECTPOINT OFF PROTECTED OFF SIZE 10 FAMILY CLASSIC OVERLINE OFF STRIKEOUT OFF UNDERLINE OFF EXPANSION REGULAR SLOPE REGULAR WEIGHT MEDIUM INVERTED OFF USERINFO NIL STYLE NIL) FORMATINFO (ARABIC )) (162 36 288 36) NIL) (HEADING NIL (HEADINGTYPE DINDE) (120 3600 492 36) NIL) (TEXT NIL NIL (72 72 468 648) NIL))) (PAGE NIL (PAPERSIZE Letter FOLIOINFO (ARABIC )) (0 0 612 792) ((FOLIO NIL (PARALOOKS (QUAD CENTERED) CHARLOOKS (SUPERSCRIPT 0 INVISIBLE OFF SELECTPOINT OFF PROTECTED OFF SIZE 10 FAMILY CLASSIC OVERLINE OFF STRIKEOUT OFF UNDERLINE OFF EXPANSION REGULAR SLOPE REGULAR WEIGHT MEDIUM INVERTED OFF USERINFO NIL STYLE NIL) FORMATINFO (ARABIC )) (162 36 288 36) NIL) (HEADING NIL (HEADINGTYPE DINDE) (120 3600 492 36) NIL) (TEXT NIL NIL (72 72 468 648) NIL))) (PAGE NIL (PAPERSIZE Letter FOLIOINFO (ARABIC )) (0 0 612 792) ((FOLIO NIL (PARALOOKS (QUAD CENTERED) CHARLOOKS (SUPERSCRIPT 0 INVISIBLE OFF SELECTPOINT OFF PROTECTED OFF SIZE 10 FAMILY CLASSIC OVERLINE OFF STRIKEOUT OFF UNDERLINE OFF EXPANSION REGULAR SLOPE REGULAR WEIGHT MEDIUM INVERTED OFF USERINFO NIL STYLE NIL) FORMATINFO (ARABIC )) (162 36 288 36) NIL) (HEADING NIL (HEADINGTYPE DINDE) (120 3600 492 36) NIL) (TEXT NIL NIL (72 72 468 648) NIL)))))(MODERN CLASSIC CLASSIC   [i ó­ÓS    C6ÖG<i(zº