Missing Values
9.1.1. Missing Values
Missing values can indicate that the respondents had a difficult time answering the posed questions. Missing values can be a problem if they represent more than 10. 242 In order to compensate for the missing values, it is possible to provide estimates for the missing data before analyzing the dataset. This approach is called mean value substitution and is used to replace the missing value by the mean of the other valid responses to that specific question. Mean value substitution can cause problems, though, and should be
239 Appendix M (disc) Output from SPSS (Generic Model) 240 Johnson, Michael D. Gustafsson, Anders (2000), p. 109 241 Ibid. 242 Malhotra, Naresh K. Birks, David F (2003), p. 431 239 Appendix M (disc) Output from SPSS (Generic Model) 240 Johnson, Michael D. Gustafsson, Anders (2000), p. 109 241 Ibid. 242 Malhotra, Naresh K. Birks, David F (2003), p. 431
Q 7 ,Q 9 ,Q 11 ,Q 12 ,Q 15 ,Q 16 ,Q 31 ,Q 37 ,Q 38 , and Q 50 go trough a mean value substitution due
to minor missing values. 244
The following questions are excluded due to higher missing values:
Q 2 and Q 3 concern local news including the amount of letters to the editor and the
information from the police report. The exclusion of these two questions must indicate that the respondents do not notice the amount of letters to the editor and the information from the police report or at least that they do not have a specific opinion
about it. This can indicate that Q 2 and Q 3 are not suitable questions for further use in
development of the model, and that these attributes are not highly influential or important to the readers, concerning their influence on loyalty.
Q 8 and Q 10 concern the variable housing including the articles about housing and interior
decoration. The exclusion of these two questions indicates that many respondents do not have an opinion about housing or maybe that they do not have articles about the subject in their WLN. These questions are removed, since a generic model is developed and if the subject is not present in all WLNs it is not optimal to use it in the generic model.
Q 20 concerns distribution and whether the paper is distributed on the publishing day or not. The high missing value in this question is most likely due to a lack of interest from the readers. They are most likely not aware of the distribution day or maybe they take household distribution for granted.
Q 26 and Q 27 concern image. The high missing values are probably due to the respondents
not understanding the questions fully or that the respondents are not aware of the WLN’s image and thereby do not have an opinion about it.
Q 32 is posed in the variable others and concerns whether the paper has a suitable amount of job advertisements or not. There are not many job advertisements in an
243 Johnson, Michael D. Gustafsson, Anders (2000), p. 110 244 Appendix M (disc) Output from SPSS (Generic Model) 243 Johnson, Michael D. Gustafsson, Anders (2000), p. 110 244 Appendix M (disc) Output from SPSS (Generic Model)
Q 34 is also a measurement variable for others and concerns the amount of “Names” in the paper. Either the WLN in the respondent’s area does not include “Names” or the respondent does not care about it. It could also be that the question includes too many different aspects and therefore the respondents have had difficulties answering the questions.
In relation to the above mentioned factors, it is also decided to completely eliminate Q25,
Q 29 ,Q 35, and Q 36, which had very high missing values. The very high percentages of missing values must be an indication of the fact that issues within image and car and garden are not as important as first initiated. The readers might also have a difficult time relating to car and garden because their WLN does not contain information about these subjects. It could indicate that people mostly discuss the information and topics of the paper but do not speak about actually reading the WLN,
which is what Q 25 and Q 29 concern. Another reason why the respondents have not
answered can be because the questions might be stated unclearly.