4 DISCUSSION
4.2 Reliability and validity
4.2.1 Reliability
The reliability of a study is, in general, related to the capability of the findings to be repeated. In quantitative research, which this dissertation mostly represents, the reliability of a study is related to the accuracy of the results, i.e. the amount of random error. This accuracy is influenced by the sample itself, its processing and finally, the interpretation of the findings. Firstly, the sample needs to be large enough and representative (not skewed). Secondly, errors should be avoided in the collection and processing the data. Thirdly, the mastery of the method is necessary in order to land in correct conclusions from the data. (Heikkilä, 2004: 30, 187).
With regard to the reliability of the current dissertation, focal factors are the quality of the data and the assumptions made during its collection and processing. The data of Articles I and II leave no room for questioning the sample size, since the samples are, in fact, entire populations in a given time frame. The sample size in Articles III and IV, then again, can be considered sufficiently large, and the data can be deemed representative of the studied cohort. As discussed in detail in the articles themselves, statistical comparisons with several other data sources indicate that the collected data
represents the entire cohort well. This view is further backed up by the repetitive nature of the findings, which indicated saturation of the data, i.e. a state in which 'no new or relevant data seem to emerge regarding a category' (Strauss & Corbin, 1990: 188), early on. This is despite the fact that the data was not picked completely randomly with regard to the geographical location of the buildings due to limitations set by the funding programme, which encompassed certain cities and towns. The collected data, nevertheless, has the vastest geographical coverage and sample size amongst similar studies so far. All in all, it is unlikely that any researcher would not land in similar findings about the flat types and panel dimensions even using a different sample.
When it comes to possible errors during the collection and processing of the data, i.e.
the quality of the data, the aforementioned data sets are also to be looked at separately.
The data of Articles I and II was pre-existing, part of the official PIS and BDR, and not collected by the author. Thus, the author could not have influenced its quality during the collection. As discussed in detail in Articles I and II, these are official data, which real estate taxation and right to use public services are based on. Therefore, the error, but the basic procedure of the compensation has been described in the article, allowing it to be repeated and validated in future studies. In Article II, then again, the need for compensating missing figures was negligible (1% of buildings).
As for the data of Articles III and IV, the collection was designed, organized and supervised but not conducted by the author. Research assistants collected the data from the archives by the means of photography. Both the source of the data and the collection method are possible causes of inaccuracy. Firstly, the archives of ARA were chosen due to its geographical coverage and accessibility, even though the archived drawings are not building permit drawings. The possibility that the drawings would have significantly changed since applying the funding from ARA seems, however, negligible, because the turnaround time was minimized during the studied era. Secondly, photography was chosen as a method to reproduce the drawings because it enabled collecting a large sample in an affordable and relatively quick way. As a result, the documented drawings were not in scale. However, as the drawings contained standardized measures (such as the width of a door or the depth of a kitchen cabinet) and most of them even had gauge lines, they could easily be stretched to scale in a
CAD program. In some cases, the photography had skewed the largest drawings from the edges, or the taken photos were not completely focused but slightly blurred. The skewedness could be corrected in the program, and the blurriness did not prevent the measurements of the dimensions. With regard to the data processing of Article IV, it is also possible that errors would have occurred while the measured dimensions were inserted to a data table, because the work was highly repetitive. Therefore, they were checked by a research assistant other than who recorded them. The possibility of this kind of error was absent in Article IV, because the method of recording was different (graphic) and the exact dimensions were not decisive for defining the flat types, even though average representations of the types were also created.
Lastly, the researcher's capability to draw correct conclusions from the findings is to be questioned. Since the main methods consisted primarily of simple statistical description, there are not many opportunities for errors that would arise from misunderstanding the limits of the method. When it comes to statistical correlations employed in Articles I and II, however, it should be noted, as always, that correlation does not equal causation, and no such conclusions have been drawn. In all, there is a reason to believe that all the findings of the research are repeatable, and thus, reliable.
4.2.2 Validity
In a valid study, the research design corresponds to the aims of the research. The research questions and indicators need to be set right with regard to the research objects. (Heikkilä, 2004: 29, 186). The aims set by the theoretical framework of the study, i.e. building stock research, which pursues to depart from case studies and to produce generalizable findings, guided the research towards an extensive approach and quantitative data and methods. The selection of data sources and the acquisition of the data were designed with quantitative methods in mind. With regard to the research objects of Article IV, i.e. recognizing repetitive flat plan designs, a purely numerical approach would not have corresponded to the aims. Therefore, a mix of statistical description and graph theory informed typological approach was employed.
In quantitative research, validity is especially related to avoiding systematic errors, which resonates with a clear understanding of the studied population and the size and representativeness of the sample. (Heikkilä, 2004: 29, 186). The lastly mentioned matters have already been discussed in the previous chapter. However, if the data of Articles I and II, which represent entire populations in a given time, are instead understood as samples in a temporal sense, a just question is in which populations can
the results, then, be generalized. This is the basic question with regard to external validity (Metsämuuronen, 2003: 35). Since the interests of the study lie rather in the future than in the past, the more specific dilemma is, thus, for how long the acquired findings describe demolition and vacancy in a valid manner. Because the findings indicate that the mortality of Finnish buildings is dynamic, i.e. changes in time, this is a highly relevant question. However, as it has been suggested that the world would soon be shifting from one economic supercycle to another, characterized, above all, by resource intelligence forced out by resource scarcity (Wilenius & Kurki, 2012: 88), the most meaningful question might, thus, not be, when the findings stop being valid, but how they can contribute to this shift.
To discuss the validity in a more conventional sense, the possibility of systematic errors in the study arises from the same factors that influence the study's reliability: the quality of the data and the choices made during the research. With regard to the data of Article I, the question is whether the data systematically omits parts of the building stock. The only group that could plausibly be omitted is that of minor cold utility buildings. Many of them escape registration in the BDR because their construction does not often require applying a permit or delivering a notification to the authorities. The skewedness of data resulting from this is, however, insignificant given the extent and characteristics of the entire building stock and the purposes of the current study.
When it comes Article II, the possibility of a systematic error is not so much related to the data itself. This is because the data encompasses residential buildings only, and they should be reliably registered due to their size and significance. In theory, the BDR could encompass some inhabitable buildings, but because the owners pay taxes based on the registered buildings, they have a strong incentive to have such buildings removed from the registry. However, the borderline between normal and problematic extents of vacancy, which had to be set by the author, is a possible source of systematic error. Its selection was based on an extensive literature review, which was practically the only way for setting it in the face of a lack of pre-existing definitions. The borderline was considered to be set at the 'safe' side, rather undermining the extent of problematic vacancy than exaggerating it. One problem in this was the different sizes of multi-family buildings. To retain the simplicity of the investigation, only one definition was given for the problematic vacancy of multi-family buildings, although different indicators could have been considered for buildings with different numbers of homes.
The fact that uses as temporary or second homes are not included in the data also influences the possibility of a systematic error arising from these definitions, since some of the homes considered empty might factually be in a kind of use. However, because the underlying motive of the research is related to ecological sustainability and
such uses tend to be irregular and thus, inefficient with regard to energy or resource use, their omission is not decisive.
With regard to Articles III and IV, the possibility of systematic errors seems highly unlikely. The geographical coverage and the size of the sample are large enough not to include omissions that could systematically distort the findings. Neither seems there to be anything in the processing method that could contribute to such an error. Rather, the possible errors stemming from it are random mistakes. Moreover, the findings of all the articles are more or less in line with the existing theory basis, which they help to sharpen and expand. This is usually considered as a sign of validity (Anttila, 2006: 512).