A more sophisticated approach is offered by advanced statistical techniques like (multiple) regression analysis. Such a technique allows the influence of individual variables to be assessed, so that the variation in rentals can be assigned to certain characteristics. Briefly, the explanatory power of a variable can be quantified via the correlation coefficient. As a by-product, it enables the characteristics to be ranked according to their importance. This helps to determine where to stratify in greater detail. Combining the most important variables, using multiple regression techniques, shows their overall explanatory power. The use of advanced statistical techniques to select important variables is considered to be an efficient way of stratifying the housing stock. In addition, regression analysis can be used directly for estimating the rentals, e.g. in the form of hedonic models. The technique is also a useful tool for estimating the average rental for strata where there are no corresponding observations in the rented sector (empty strata).
A further advantage of selecting stratification criteria on the basis of an advanced statistical technique is that it avoids the need to prescribe uniform criteria for all Member States. To obtain a comparable result, it is sufficient to establish a ranking of the most important criteria in each Member States and to stipulate the required overall level of explanatory power. Obviously, such a regression analysis depends largely on the available statistical information. However, even in a situation of restricted statistical information, this could be an incentive for future improvements.
Given that the information about the different variables affecting rentals mainly depends on the development of basic statistics, the possibility of using advanced statistical techniques may be restricted at present. Therefore a standard method is recommended, i.e. Member States shall apply all significant criteria as derived from tabular analyses. As a minimum, the size, the location and at least one other important feature of a dwelling have to be used to stratify the housing stock; this stratification should produce a minimum of 30 cells. The breakdown of the housing stock has to be meaningful and representative of the total stock of dwellings. An advanced statistical technique may be used to determine the important explanatory variable(s) for selecting the strata.
In practice, however, a Member State may prefer to use fewer variables, or variables other than those prescribed by the standard method. This is acceptable, as long as a (multiple) regression analysis proves an acceptable level of explanatory power. To guarantee comparable results, a correlation coefficient of at least 70 % is recommended as a threshold. This threshold value would be acceptable in the context of a large sample, with zero and cheap rentals as well as outliers having been removed.
Principle 3
:
Member States shall use tabular analyses or statistical techniques to derive significant stratification criteria. As a minimum, the size, the location and at least one other important feature of a dwelling have to be used. A minimum of 30 cells are to be produced and at least three size classes and two types of location shall be distinguished. The use of fewer or other variables is acceptable if it has been proved previously that the (multiple) correlation coefficient reaches 70 %.