# Cluster Analysis - Round Two: Cluster household and housing characteristics

 Page 60/60 Date 30.04.2018 Size 13.27 Mb. #42625

#### Cluster Analysis - Round Two: Cluster household and housing characteristics

The following tables provide data on the household and housing characteristics of the four clusters. The data is either in the form of percentages of households in each cluster represented by an option for different variables (e.g. Single Person, or retired) or descriptive statistics for ordinal characteristics (e.g. house price, number of bedrooms). Each table has a limited text describing the key variations in the tables, further explanation of the key characteristics can be found in chapter eight.

Table 3.1: Household type by cluster

 A B C D Simple household typology Single Person 36% 26% 12% 57% Couple (no children) 33% 47% 50% 21% Couple (with children) 14% 19% 29% 7% Lone Parent 11% 5% 4% 0% Extended Household 6% 3% 4% 14%

Cluster C has the highest proportion of couple households (with or without children), whilst Cluster D has much less. Lone Parents are more likely to be found in Cluster A than any other, whilst Extended Households were most likely to be found in Cluster D.

Table 3.2: Education of respondent by cluster

 A B C D Education of respondent Postgraduate/Professional Qualification 42% 37% 35% 31% Degree or degree equivalent 19% 34% 24% 0% Higher education below degree 11% 6% 12% 15% A Levels / NVQ Level 3 5% 11% 8% 8% GCSEs / NVQ level 2 7% 5% 12% 0% NVQ Level 1 0% 0% 1% 0% No qualifications 4% 2% 2% 15% Other 5% 2% 3% 8% No response 6% 3% 4% 23%

Cluster B has the highest proportion of household respondents with a degree or postgraduate degree (71%), whilst cluster D has the smallest proportion (though this cluster had a high non response rate)

Table 3.3: Working status of respondent by cluster

 A B C D Working status of respondent Full-time employment 42% 72% 46% 36% Part-time employment 13% 15% 22% 14% Self-employed 6% 3% 9% 0% Unemployed 2% 0% 1% 0% Retired 32% 3% 14% 50% Full-time student (16+years) 1% 2% 2% 0% School / preschool / nursery 1% 0% 0% 0% Looking after home / family 1% 3% 4% 0% Permanently sick / disabled 0% 1% 2% 0% Other 2% 1% 2% 0%

Clusters A and D had high proportions of Retired respondents, whilst Cluster B had by far the highest proportion of respondents in employment (full-time employment, part-time employment or self-employed).

Table 3.4: House price by cluster

 A B C D House Price Number 162 175 113 14 Missing 0 0 0 0 Mean £203,517 £163,115 £221,499 £138,596 Median £162,475 £134,950 £199,950 £99,950 Std. Deviation £140,739 £98,153 £117,840 £93,432 Minimum £50,000 £58,000 £70,000 £52,000 Maximum £1,175,000 £750,000 £732,500 £389,950 Percentile 25 £118,750 £108,000 £135,750 £81,375 Percentile 50 £162,475 £134,950 £199,950 £99,950 Percentile 75 £235,000 £186,000 £255,019 £183,000

Cluster D had the lowest median house price (and lower and upper quartile prices), whilst Cluster C had the highest house price lower quartile, median and upper quartile house prices. Cluster A has a higher house price profile than Cluster C.

Table 3.5: Household income by cluster

 A B C D Household Income Number 139 162 98 11 Missing 23 13 15 3 Mean 9.79 10.43 11.88 6.18 Median £35-40,000 £35-40,000 £50-£60,000 £20-22,500 Minimum £5,000 or under £5,000 or under £5-10,000 £5,000 or under Maximum £80,000 or over £80,000 or over £80,000 or over £40-50,000 Percentile 25 £20-22,500 £25-27,500 £35-40,000 £5-10,000 Percentile 50 £35-40,000 £35-40,000 £50-£60,000 £20-22,500 Percentile 75 £50-£60,000 £50-£60,000 £60-70,000 £30-35,000

Cluster D had the lowest median income (and lower and upper quartile incomes), whilst Cluster D had the highest household lower, median and upper incomes. Clusters A and B had very similar income profiles.

Table 3.6: Number of people in household by cluster

 A B C D Number of people in household Number 162 175 113 14 Missing 0 0 0 0 Mean 2.0 2.1 2.5 1.6 Median 2 2 2 1 Std. Deviation 1.0 1.0 1.2 0.8 Minimum 1 1 1 1 Maximum 5 5 6 3 Percentile 25 1 1 2 1 Percentile 50 2 2 2 1 Percentile 75 2 2 3 2

Cluster C had the highest mean number of people in the household (and highest lower and upper quartiles), whilst Cluster D had the lowest numbers. These figures correspond closely to the simple household typology and the number of households with children (in Cluster C predominantly) and Single Person households (Cluster D).

Table 3.7: Number of bedrooms by cluster

 A B C D Number of bedrooms Number 156 174 111 14 Missing 6 1 2 0 Mean 3.7 3.9 4.2 3.6 Median 4 4 4 4 Std. Deviation 1.0 0.9 1.0 0.7 Minimum 1 1 2 2 Maximum 6 6 8 5 Percentile 25 3 3 4 3 Percentile 50 4 4 4 4 Percentile 75 4 4 5 4

There is not a large amount of variation in the number of bedrooms between clusters. Cluster D has the fewest according to the mean, but there is no variation in the medians.

Table 3.8: Number of bathrooms by cluster

 A B C D Number of bathrooms Number 156 174 111 14 Missing 6 1 2 0 Mean 1.6 1.5 1.8 1.4 Median 1 1 2 1 Std. Deviation 0.8 0.8 0.9 0.6 Minimum 1 1 1 1 Maximum 4 5 5 3 Percentile 25 1 1 1 1 Percentile 50 1 1 2 1 Percentile 75 2 2 2 2

Cluster D is smaller than other clusters; using the measure of bathrooms it has the fewest according to the mean and maximum values.

The net agree scores are discussed in the main body of the thesis (chapter seven) and are therefore not described here.

Table 3.9: Net agree score for awareness before searching by cluster
 NET AGREE A B C D ALL Knew Neighbourhood 46% 27% 46% 36% 38% Neighbourhood Aware 49% 65% 71% 38% 61% Location Unimportant 86% 72% 82% -100% 74% Exact Type Aware 38% 18% 31% 33% 28% Type Aware 51% 57% 59% 67% 56% Type Unimportant 46% 27% 46% 36% 38% Exact size 46% 27% 46% 36% 38% Size Aware 46% 27% 46% 36% 38% Size Unimportant 46% 27% 46% 36% 38%

Table 3.10: Net agree score for attitude before searching by cluster
 A B C D TOTAL Constantly Considered Moving -81% -26% -12% -33% -42% Interested Better Home -42% 35% 50% -27% 11% No Better Home Available 24% 21% 46% 17% 28% Too Much Effort 35% 48% 57% 18% 45% Satisfied Previous -31% 22% 18% 36% 3% Satisfied Finance -49% -23% -59% -27% -41% Occasionally Considered 5% 35% 46% 27% 27% Didn't Consider Until Event -54% -11% 7% -82% -23%

Table 3.11: Net agree score on motivations (1) by cluster
 A B C D TOTAL Economic Change -20% -11% -41% -18% -22% Family Change 42% -10% -2% 20% 10% Design Or Size Dissatisfaction -36% -16% 29% 20% -11% Area Dissatisfaction -57% -35% -17% 30% -36% Financial Dissatisfaction -70% -34% -68% -40% -55%

Table 3.12: Net agree score on motivations (2) by cluster
 A B C D TOTAL Increase Wealth -49% -24% -43% -42% -37% Increase Social Status -81% -51% -66% -50% -65% Increase Comfort 14% 62% 71% 36% 47% Increase Stimulation -33% 9% 5% -9% -7% Enable Personality -23% 7% -12% 64% -6% Closer Friends -13% -28% -35% 27% -23% Good Social Exposure -46% -26% -39% 36% -35%

Table 3.13: Net agree score on motivations (3) by cluster
 A B C D TOTAL Affordable 70% 87% 88% 83% 81% Deposit -19% 63% -5% -27% 16% Interest Rate -24% 60% 24% -9% 22% Flexible lending conditions -33% 42% 16% -27% 9% Low inflation -39% 35% 13% -20% 4% Price rise -42% 32% -11% -9% -5% Interest rise -48% 21% -10% -17% -11% Stamp duty -43% 22% -36% 9% -14% Rent rise -70% -24% -77% -40% -53% Pay rise -73% -7% -61% -27% -43% Birth -69% -44% -56% -9% -55% Relationship Change -33% -37% -49% -60% -39% Relocation Job -60% -36% -72% -45% -53% New Job -59% -40% -85% -45% -58% Difficult to rent -73% -71% -93% -82% -77% Space pressure -50% 10% 17% 17% -8% Home physical pressure -12% 12% 41% 9% 11% Home finance pressure -53% -31% -73% -27% -49% Not afford home -68% -80% -91% -40% -78% Specific property available -38% -34% -20% 0% -31%

Table 3.14: Net agree score on search time pressures by cluster
 A B C D TOTAL Length Time Pressure -24% -38% -60% -36% -39% Search Time Pressure -16% -13% -50% -38% -24% Time Increase -16% 6% -25% -23% -10%

Table 3.15: Net important score for information source by cluster

 A B C D TOTAL Personal Knowledge 96% 78% 93% 100% 89% Friends 30% 37% 11% 50% 29% Newspapers -44% -34% -2% -33% -29% Agent Window -19% -6% -2% 0% -9% Agent Person -27% -14% 5% -23% -14% Agent Website 25% 49% 76% -9% 46% Internet Property 42% 78% 89% 17% 67% Internet Area -17% 38% 24% -27% 15%

Table 3.16: Very often and often answers for information source by cluster
 A B C D TOTAL Personal Knowledge 83% 71% 96% 86% 82% Friends 24% 53% 16% 50% 34% Newspapers -30% -18% 15% -17% -14% Agent Window -6% -2% 15% 27% 1% Agent Person -25% -12% 15% -17% -10% Agent Website 29% 64% 78% 33% 55% Internet Property 50% 79% 85% 9% 69% Internet Area -14% 38% 15% -27% 13%

Table 3.17: Percentage of households who altered their hopes between stages by cluster
 A B C D TOTAL % altered between first visit and search 16% 24% 20% 50% 21% % altered between first offer and first visit 17% 23% 19% 43% 21% % altered between move and first viewing 20% 22% 24% 43% 22%

Table 3.18: Net agree score on search experience (1) by cluster
 A B C D TOTAL Knew Found Quick 49% 54% 35% 33% 47% Knew Found Eventually -28% 1% 5% -17% -8% Knew Change Price -29% 5% -12% -27% -11% Knew Change Size Or Area -41% -21% -32% -18% -30% Uncertain Now Happy -36% -20% -46% 18% -31% Uncertain Now Not Entirely Happy -86% -77% -86% -73% -82%

Table 3.19: Properties visited and placed an offer on by cluster
 Ward Method N Minimum Maximum Mean Std. Deviation A S.3.b_PropPhysicallyVisit 155 1 130 8.21 13.002 S.3.c_PropOffers 156 0 5 1.52 0.876 B S.3.b_PropPhysicallyVisit 175 0 50 8.99 8.96 S.3.c_PropOffers 175 1 12 1.62 1.178 C S.3.b_PropPhysicallyVisit 113 1 50 8.34 8.043 S.3.c_PropOffers 113 1 4 1.68 0.794 D S.3.b_PropPhysicallyVisit 14 1 30 5.57 7.325 S.3.c_PropOffers 14 0 3 1.36 0.745

Table 3.20: Search time and length by cluster
 Ward Method First Consider and First Search First Consider and First Viewing First Consider and First Offer First Consider and Move A No. 154 153 154 153 Missing 8 9 8 9 Mean 5.73 7.33 10.39 15.29 Median 1 2 5 9 Std. Deviation 18.224 22.125 20.847 21.986 Minimum -11 -61 -10 -6 Maximum 204 221 221 229 Percentile 25 0 0.5 1 5 Percentile 50 1 2 5 9 Percentile 75 5.25 10 13 20 B No. 170 171 169 172 Missing 5 4 6 3 Mean 1.78 3.95 5.66 10.31 Median 0 2 4 7.5 Std. Deviation 3.464 5.522 5.944 7.86 Minimum -7 -3 -2 -5 Maximum 19 39 33 48 Percentile 25 0 1 2 5.25 Percentile 50 0 2 4 7.5 Percentile 75 2 5 8 13 C No. 109 108 110 111 Missing 4 5 3 2 Mean 4.11 7.5 9.81 14.26 Median 0 2 5 10 Std. Deviation 13.657 15.882 16.05 21.038 Minimum -3 -3 0 -117 Maximum 108 113 113 115 Percentile 25 0 1 2 6 Percentile 50 0 2 5 10 Percentile 75 2.5 7 10 16 D No. 13 12 12 13 Missing 1 2 2 1 Mean -4.62 6.42 6 10.77 Median 0 3.5 3.5 9 Std. Deviation 24.147 6.96 8.068 7.224 Minimum -84 1 -8 2 Maximum 12 24 24 28 Percentile 25 0 1 1.25 5 Percentile 50 0 3.5 3.5 9 Percentile 75 2.5 11.25 12 16.5

Table 3.21 Previous location of dwelling (inside or outside Sheffield)
 Cluster Outside of Sheffield (Postcode) Within Sheffield move A 20% 80% B 9% 91% C 7% 93% D 25% 75% Total 13% 87%

Chart 1: Cluster geographic distribution

1 These households are referred to as owner-occupiers in the thesis, i.e. they are identified by their outcome tenure. Whilst some of these households were owner-occupiers in the previous dwelling others were previously renting (either privately, from the council or from a registered social landlord) and yet others were also living as part of another household (for example they were living with parents).

2 There were 5,183 dwelling transactions in 2010 in Sheffield, of which 4,771 were second hand and 412 were new build according to HMLR data.

3 In 2011, 58.3% of all households in Sheffield lived in owner-occupied properties, compared to 15.6% in the private rental sector and 24.8% in the social rented sector. The variation in the proportion of households in owner-occupation was significant between market areas in the city, ranging from 12.8% in the City Centre to 78.1% in the Rural Upper Don Valley area of the city (ONS, 2011).

4 Economics is often criticized for not recgonising the normative position it embraces (Berg, 2003; Davis, 2006a) or the normative positions of actors in those theories (Etzioni, 1988)

5 There is currently a debate between theorists arguing against the concept of schools in economic thought and those for their existence. There is legitimacy to the dual concern that differentiation by school type masks variation between economists categorized within schools and overlaps in their theories between schools, and also erect barriers to communication between economists. This deconstructionist approach can be extended to question whether economics as a field is stable over time (Davis, 2006b), and therefore whether it is possible to categorise distinct schools if the field is changing. However, avoiding this level of regress, schools also represent heuristics for exploring key issues in conceptualizing markets and the behaviour of actors in those markets, and are used in this thesis as a way of segmenting a large and diffuse literature.

6 There are also almost as many variations of definition of significant schools as there are economists. For example, Brue and Grant (2012) in a more susbstantive overview of the evolution of economics consider: The Mercantilist School, The Physiocratic School, The Classical School, Marxian Socialism, The German Historical School, The Marginalist School, The Neoclassical School, The Institutionalist School, the Keynesian School and the Chicago School. Samuels, Biddle and Davis (2003), also include Ancient and Medieval Economics, Post-Ricardian British Economics, Utopian Economics and Feminist Economics. Others, such as Tsoulfidis (2010) segment economic thought by principal authors (e.g. Smith, Ricardo, Marx and Keynes) rather than defined schools. In short, there is no agreed definition of schools of thought.

7 There is an element of overlap between the schools. BE has some similarities to IE and some adherents also link it to NCE. As well as the schools discussed here BE has also drawn inspiration from other economic perspectives which are not expanded here including, Schumpeterian, game theory and public choice theory (Coats, 2014).

8 Neoclassical economics is not a uniform field and its use as a label may even be unhelpful in masking the widespread variation (see Colander, 2000). Even some of the most famous NCE economists have questioned some of the foundational tenets. For example Adam Smith (1759) wrote: “How selfish soever man may be supposed, there are evidently some principles in his nature, which interest him in the fortune of others, and render their happiness necessary to him, though he derives nothing from it except the pleasure of seeing it” (Smith [1759] 2002, 11).

9 Some commentators make a distinction between NIE and Neo-Institutional Economics (NeoIE). They argue that NeoIE does not break with the rational tenets of NCE (e.g. Eggertsson, 1990), grafting on insights about institutions to mainstream theory, but NIE does (Furubotn and Richter, 2008; Arvanitidis, 2015).

10 The term ‘school’ is used here, as it has been used to distinguish between NCE, AE, ME and IE. There is some disagreement about whether BE can be considered as school given the extensive variation between economists defining themselves in relation to BE. Tomer (2007) argues that there are substantive differences between strands of BE, and these strands may more appropriately be called schools rather than BE as a school, however given the overall consensus in BE that the great problem is NCE’s incapacity to accurately describe or predict behaviour, in this sense BE may be called a school.

11 There are a number of papers which are particularly complex to locate in this bfuraction between old and new. Lindenberg (1990) for example argues that sociology promises a more fruitful line of inquiry than psychology, but ends by augmenting homo sociologicus with homo economicus, and thus retaining some of the disputed tenets from both OBE and NBE in homo socio-economicus.

12 The focus on psychology in NCE has caused some commentators to refer to it as ‘Psychological Economics’ rather than NCE, but the substantive distinctions identified remain the same as a strand of research within BE (e.g. Tomer, 2007)

13 Simon’s work on bounded rationality was never fully integrated into a precise definition of economic humans. Issues of the amount of information, social relations and emotions were frequently highlighted, but not systematically integrated (Lindenberg, 2001).

14 Financial search is often considered as part of the housing search process, as many households have to both sell a previous home and agree a loan in order to purchase a new home. It is, however, not included in the above definition because the search for a dwelling and the search for finance take place in different market places (although they impact on each other). Homebuyers finance search is heterogeneous (Duffy and Roche, 2005), adding to the complexity of the variation between households purchase behaviour.

15 It is not unusual to see properties, which were originally designed to be uniform, that have been substantially altered. Le Corbusier famously found the changes occupiers made to their properties a frustration to his identikit designs.

16 Although it is not a part of this study, given the significant expenses, finance is often required in order to undertake the purchase, increasing the complexity as a joint good (Gibb, 2009). The availability and cost of debt finance influences the price of housing, as for many households both finance and the dwelling are purchased simultaneously. Likewise, Tenure is not considered in this study, however The possibility of living in different tenure accommodation adds a further layer of complexity to the housing market, as in theory, different tenures compete against each other for residents’ selection according to the households’ beliefs in the benefits (Bogardus Drew, 2014). A comparison therefore between renting and owning a dwelling may also play a part in housing search decisions and may add to the complexity of both financial calculations and projecting lifestyles. Whilst this is a major complexity in many circumstances, it is not considered as part of this thesis.

17 There is some variation in the relationship between consumption and investment at the national scale, suggesting that institutional considerations are significant for understanding the relationship (see Henderson and Ioannides, 1983 and the comparative work of Arrondel and Lefebvre, 2001)

18 The housing search can be considered from a wide number of perspective, and determining the unit of analysis is not straightforward. The rationale for using the term household, rather than individual is developed further in Chapter Five.

19 There is a caveat to this claim. This view does not consider time, and the delays between a property becoming available and marketed online. In slow markets this may be of little concern, but it is feasible that in hot markets the property may be transacted before the details have been submitted online. This may be more likely in the rental market than in the freehold market, where time may be of greater significance.

20 Sheila Dow (2000) argues that the difference between closed and open systems is the key division between orthodox and heterodox economics (respectively). Orthodox systems rely on the knowability of all relevant relationships and variables, whilst heterodox systems suggest that these cannot be full known as they evolve.

21 The full criteria, and responses in this research instance are set out in Appendix A

22 Appendix B contains the key references for each of the variables used in the survey

23 These are: categorical; dates; postcodes; Likert; binary; text; number; ordinal; and interval

24 There are competing opinions about the appropriate number of points on a Likert item (Likert Scales are multiple questions, items are single questions). Dawes (2008) found that in marketing research there was no significant difference between respondents (re-scaled) answers on five and seven point scales, although forced choice scales had the effect of lowering mean scores. Cummins and Gullone (2000) likewise argue that there is no reliability issue, although for some measures a seven-point scale can increase sensitivity. Ease of completion for the survey was a key issue (Fowler, 2009) especially given the size, therefore the decision to use a five point scale was taken to minimize the complexity of each question (on the basis that it is easier to decide intuitively between Strongly Agree and Agree than four points on a scale encompassing Strongly Agree and Agree).

25 Questions of household income may be understood if it is not clear whether the income is before or after tax, or what the period of income should cover (Peterson, 2000). The question was carefully phrased therefore to be clear that the answer should relate to annual and gross income.

26 This method may, in some cases overestimate question understanding as respondents simply repeat the phrases used in the questions (Foddy, 1998). This was avoided by asking the participants to elaborate on the meaning of the questions in detail and therefore test their cognition.

27 It is possible that the ownership of a dwelling changed hands, but the transaction was not registered with the HMLR. This informal market activity cannot accurately be quantified, or the addresses identified. Therefore using HMLR data for the sample population may underrepresent the target population, but this is the most suitable solution for identifying addresses.

28 There is competing evidence about the impact of monetary rewards on response rates. Yammarino (1991) and Edwards et al (2002) argue that payments are likely to increase rates, although Fowler (2009) provides evidence from a number of studies, which suggest only cash advance payments have an impact.

29 Further research is needed to test the results of this research in other market conditions before any assumptions of the correlation between housing or household characteristics found in this research can be applied to housing search behaviour in other studies.

30 A significant number of other residential segmentation studies use PCA and CA in a single study, but simply use PCA to reduce the number of original variables in the CA, by selecting those with the highest eigenvalues (e.g. Maclennan and Tu, 1996), although this accounts for a significantly reduced amount of the variation.

31 This section draws together accounts provided in Joliffe (2002, p.6-9), entitled A Brief History of Principal Component Analysis, in the preface to Jackson (1991), and in Brubaker’s PhD (2009).

32 There are two options for ordinal data in CATPCA: Spline Ordinal and Ordinal. Spline ordinal requires a chosen degree for the line and interior knots, ordinal does not require this and produces a closer fitting line, but may be less smooth (Meulman and Heiser, 2001).

33 The DCLG (2013) household typology for household projections has two layers. The first includes: One Person, One family and no others, A couple and one or more other adults, Lone parent and one or more other adults, and Other households. The second layer differentiates between couple and single parent households and the number of children in the household. The typology used in this research includes the first category distinctions in the second layer (lone parent or couples), but not the number of children (although this is recorded in the survey) and conflates the multiple adult households where they are not a couple

34 As with the discussion on search processes in chapter three, there is no suggestion that households will move through each of these stages sequentially, but the model is provided as a mechanism to show where variation may occur.

35 Information about the location of dwellings may relate to a large number of variables. Dunning et al. (forthcoming) show that estate agents are playing a role in refining households’ perceptions of neighbourhoods including information about schools, employment and recreation. There is a growing literature on the role of information sources on people’s perceptions of neighbourhoods and impact on housing markets (Burrows et al., 2005).

36 It is important to note here that the similarity between revealed household behaviour from the survey and some aspects of the NCE theory of human behaviour in the form of utility maximization does not indicate that the totality of homo economicus is likely or even possible. Evidence that some households constantly consider moving does not suggest that they know all opportunities available, or that they are able to compute the range of attributes and housing market information that is necessary to fulfill the requirements of homo economicus.

37 NB: households may have moved for multiple reasons and the survey was designed to collect respondents’ answers to each variable, therefore the answers cannot simply be summed. For example a household may have strongly agreed that both a new job and a pay rise were important factors in their decision to move.

38 Dunning et al. (forthcoming) have shown that estate agents recognize that they now play a limited role in the initial formation of households’ views of the market and are now working harder to influence these views.

39 Households are clustered based on their similar (proximity) scores on the CATPCA components. Through the CA process in SPSS a new variable is created which identifies which cluster each case (household) belongs to.

40 The first CATPCA component loadings are not tabulated here as these results were not then used for further analysis, they are however presented in full in appendix E

41 Other loading values may be used to identify the key variables. 0.3 is often used by researchers (Bryman and Cramer, 2005), but there is no mathematically robust way of determining what the value should be, therefore 0.5 was selected as this gave a clearer indication of key variables than relaxing it to 0.3. Given the purpose here is descriptive some value below 0.5 are included on components with lower loading scores.

42 Some caution should be heeded for Cluster D for these variables as the response numbers were particularly low for these questions with a mixture of non-response and non-readable answers.

43 There is a significant literature on comparative policy studies in housing (e.g. World Bank, 1993), but very little on the comparisons between search behaviours (Dieleman, 2001).