The results of the stepwise MLR process are presented in Table 4.The model results show the selection variable parameter estimatesand odds ratios for the IDM and DBA architectures with EDW as thereference architecture. The table reveals that strategic view, resourceconstraints, and perceived ability influence the selection of a datawarehouse architecture when comparing the IDM and EDW architectures.Also, a strategic view was significant at pb0.05 in influencingthe odds of selecting the DBA architecture when compared to EDW.Specifically, the odds of selecting a DBA architecture when comparedto selecting an EDW architecture decreased by 19.5% as strategic viewincreased by 1 unit. The analysis indicated that perceived ability andresource constraints did not influence the chance of selecting a DBAwhen compared to the EDW architecture.The multinomial model was also assessed for its discriminatingpower. Assessing a multinomial model's predictive power is problematicwhen most of the responses are in a single category [53].Morespecifically, for the choice of data warehouse architectures, it isdifficult to create a model that outperforms using the dominantcategory (EDW was used 61% of the time) as the prediction for everycase. In response, a two-step assessment approach was used.First, the model's predictive power was tested against a “naïvemodel.” With a naïve model, every response category (i.e., the variousarchitectures) is assumed to be equally likely. Applying the naïvemodel, one of the three architectures is randomly selected, with any ofthe three architectures as likely to be predicted as the others. Withthis approach, the expected value is that the prediction is correct33.3% of the time. For comparison purposes, when the multinomialmodel was used, it predicted the correct architecture 62.8% of thetime.The second approach was to compare the multinomial model'spredictions for IDM and DBA against the percentages of responses inthose architecture categories. In other words, can the model predictwhich architecture would be used when the largest category isremoved? There were 104 DBA and 44 IDM responses for thesearchitectures, and the model predicted the correct architecture 76.8%of the time.