This paper focuses on four measures of household space:
Dwellings per person: total dwellings / total persons (including non-household residents)
Overcrowding: percentage of households with more than 1 person per room
Rooms per person: total rooms / total persons (including non-household residents)
Average household size: mean average number of household residents per household
Dwellings per person is simple to compute given only data on dwellings and people (here total persons) and provides a useful headline figure, especially in relation to assessments of house building / availability.
Overcrowding explicitly relates rooms within a given household to the number of occupants (thus, it can only be calculated if data on individual households and rooms in each household are available). This property means that it has been a key focus of research and also policy, as noted in the introduction.
Rooms per person builds on dwellings per person, but properly reflects (albeit indirectly) dwelling spaces.
Average household size has a rather different focus to the previous three measures as it need not imply limited housing space if many rooms (relative to numbers of occupants) are available. It is an important accompaniment to the other three measures.
Results using each of these measures is assessed in turn, with a focus on trends through time in neighbourhoods across England. Relationships between the measures are discussed later in the paper.
Dwellings and People
Dwellings were not counted in the 1981 Census (or the 1971 Census) but were included in the 1991 Census following pressure from central government departments, local and health authorities (Thompson, 1995). Estimating dwellings counts from the available household data was judged to be inadequate and the level of uncertainty was considered too high. Table 2 shows the number of dwellings and persons in each of the nine regions of England in 1991, 2001, and 2011. The total numbers of dwellings and persons for each year for England as a whole are also given. Total persons rather than persons in households are used for consistency with analyses for the post 2011 Census period (see Section 5), for which small area household resident counts are not available. The number of households, rather than individuals, would (if known) better enable assessment of how far there is pent up demand associated with households sharing dwellings, but would not by itself shed light on problems of constrained household formation. Household formation is affected by the availability of suitable housing, and when housing supply matches demand then household size tends to fall (partly due to people demanding more space per person as incomes rise, and partly due to demographic trends leading to rising numbers of single-person or childless households) (see Bramley et al, 2010).
In England as a whole the number of dwellings grew 8% between 1991 and 2001, while the number of people grew 4.5%—indicating a growing amount of space per person. Between 2001 and 2011 both dwellings and people grew by 8%. London stands out as the only region where growth in people outpaced growth in dwellings in both decades. As will be shown, however, in every other region there were some neighbourhoods where housing growth failed to keep up with growth in the population. The number of dwellings per person in each Census year (derived from Table 2) is shown in the Appendix Table 7. Only in London did the number of dwellings per person decrease over the period 1991 to 2011. It is worth nothing that the coefficient of determination for the total number of persons and total household residents in 2011 is 0.808; there is a relatively small number of LSOAs within which the total number of people is much larger than the number of residents in households (due to student halls of residence and other communal establishments) and so the general geographical trends are robust.
Between 1971 and 2011 there were population decreases in many urban areas of England including Newcastle, Manchester, Liverpool, Birmingham and central London, with increases in the outskirts of London and other areas across, most notably, the south east of England. There is evidence for counter-urbanisation from areas with higher population densities between 1971 and 1981, followed by a gradual increase in urbanisation from 1981 to 2001, and a large increase from 2001 to 2011 (see Lloyd et al., 2017, and also Champion (Ed.) 1989, and Champion, 2008, on more recent trends). Did housing provision keep up with this renewed urban population growth? Fig. 1 shows the ratio of dwellings to persons in (a) 2011 and (b) the change between 2001 and 2011 (in practice, 2011 counts minus 2001 counts to give positive values for increases and negative values for decreases). Note that in the maps blue colour ranges are used for single timepoints and a spectral range for maps of changes and for maps of years with the maximum value of a given measure. In 2011, the numbers of dwellings per person was consistently smaller in urban areas than elsewhere, although in the most expensive parts of central London (largely in the boroughs of Westminster and Kensington and Chelsea) there are relatively large numbers of dwellings per person due to high rates of second home ownership. The largest decreases in dwellings per person between 2001 and 2011 were in outer London and in other urban areas including Birmingham.
Figure 2 shows change in dwellings per person between 2001 and 2011 against initial (2001) population density. In every region, more densely populated neighbourhoods tended to see the smallest increase or the greatest decrease in dwellings per person, consistent with a pattern of housing supply failing to keep up with resurgent population growth in the centres of towns and cities. Figure 2 shows that the relationship between change and initial density was stronger in regions other than London, where typical densities are higher than in other regions and where neighbourhoods throughout the density distribution tended to see a fall in dwellings per person (although Fig. 1B above shows that there were increases in dwellings per person in a few central London neighbourhoods).
Overcrowding declined markedly across England between 1971 and 1981 (see Table 3) before levelling off in most regions and then increasing in London (starting in the 1990s) and the wider south east and midlands (in the 2000s), corresponding to evidence for counter-urbanisation from areas with higher population densities between 1971 and 1981, followed by a gradual increase in urbanisation from 1981 to 2001, and a large increase from 2001 to 2011 (see Gleeson, 2015a). The overcrowding rate had been falling continuously since the 1930s (see chart 5.12 in Gleeson, 2015b), and while housing supply in the UK was acknowledged to be relatively unresponsive to demand (see Barker, 2004), population growth from around the mid-1990’s onwards was generally higher than expected (see OBR, 2017) while housing supply was also hit by the (mostly unanticipated) global financial crisis.
Overcrowding follows similar trends to dwellings per person (Table 2) and changes in the two variables over the period 2001 to 2011 are strongly correlated (the coefficient of determination is 0.82). There are some differences at a regional level – whereas overcrowding in the south east increased between 2001 and 2011, the overall numbers of dwellings per person grew, suggesting a growing inequality of space per person (see Tunstall, 2015).
The region-level summaries again mask considerable variation, and local-level data are essential for understanding the changing spatial distributions of overcrowding. In 2011, overcrowding rates were high in outer London and in Birmingham, as well as in some other urban areas (see Fig. 3A). By far the most marked change between 1971 and 2011 was in outer London where there were large increases (Fig. 3B). Taking the period 2001 to 2011 specifically, large increases in overcrowding are evidenced in London but also in Birmingham (Fig. 3C). An increase in overcrowding in London relative to the rest of the country is observed by Gregory et al., (2001; where overcrowding is defined as > 1.5 persons per room) over a much longer time period — 1901 to 1991. Gregory et al. show that, while overcrowding rates fell sharply between 1901 and 1991, there was an increase in the inequality of overcrowding rates at the national level, and the rank of London’s overcrowding rate among regions rose over time. Figure 3D shows the Census year in which overcrowding was highest in each LSOA – in most areas 1971 saw the highest overcrowding, but a band of LSOAs with their largest values in 2011 is apparent in outer London. At the national level overcrowding fell over the course of the second half of the twentieth century (see also Gleeson, 2015b, Chart 5.12 which demonstrates this trend for areas within London over the period 1931 to 2011), and since the 1990s has only increased in pockets, so given that measurement starts in 1971, that would be expected to be the most common maximum.
The increasingly strong contrast in overcrowding between urban areas and elsewhere corresponds to an increase in housing space inequalities. One way to represent these inequalities is to use an index of segregation. Here, the index of dissimilarity (D; see Lloyd, 2015 for an analysis of multiple demographic and socio-economic variables including housing tenure) is used to measure the geographical ‘spread’ of overcrowded households relative to those which are not overcrowded. The index would take a value of zero if all LSOAs had the same proportions of households in both categories (for example, 80% not overcrowded and 20% overcrowded households in each LSOA – a completely even spread). The maximum possible value is one – this would indicate a case where all LSOAs comprise 100% non-overcrowded households or 100% overcrowded households (a completely uneven spread). Table 4 shows the index value for each Census year. In line with the maps, the figures show a decade-on-decade increase in unevenness, signalling an increased polarisation between areas with high and low rates of overcrowding. Note that the segregation measure requires counts of two groups, such as overcrowded household and non-overcrowded households. Overcrowding is the only measure of household space which can be dichotomised in this way.
The largest increases in overcrowding between 2001 and 2011 tended to be in areas with increases in the proportions of people from Black, Asian and Minority Ethnic (BAME) groups; there is a strong negative correlation between White British % and Overcrowding % in 2011 (r squared = 0.71), and a moderate correlation between the changes in these two variables between 2001 and 2011 (r squared = 0.21). BAME groups are more likely to live in the private rented and social rented sectors, which absorbed all of the increase in overcrowding in this period (MHCLG, 2018). For more on this topic, see Finney and Harries (2015), and Johnston et al. (2016), who focus on London.
To more fully understand the determinants of change in overcrowding, a regression model was constructed which seeks to explain changes in overcrowding between 1971 and 2011. The selection of independent variables was guided by the previous analyses, the literature, and via stepwise selection. The results (Table 5) suggest that overcrowding generally increased when:
The total population rose
Places were less overcrowded in 1971
Average household size rose
The total number of households fell (likely driven by the greatest increases in the number of households being seen in places that were relatively low density in 1971)
The private rented sector grew
The social rented sector shrank
The White British share of the population was low in 2011.
The overcrowding in 1971 coefficient sign reflects the marked increase of overcrowding in areas which had previously low levels—outer London is the most obvious case. Changes in private renting and social renting are strongly correlated but tests for multicollinearity (including the condition number and variation inflation factors; see Belsley et al., 1980) do not suggest that either variable should be removed from the model.
The analysis in Bramley et al. (2010) demonstrates that higher levels of overall new housing supply and the supply of social housing, in particular, would tend to increase household formation (thereby lowering average household size) and reduce overcrowding (see pp. 10–11). The relatively low levels of housebuilding in England and the shrinkage of the social housing sector (reflected in the model outputs) are therefore part of the explanation for the observed patterns.
The spatial trends in overcrowding between 1971 and 2011 are complex. One way of better understanding the differences and commonalities in trajectories between areas is to employ a classification of these trajectories. The R (programming environment; R Core Team, 2017) package kml (Genolini et al., 2015) implements a version of k-means clustering adapted to time-series data such as those analysed here. In this case, the overcrowding percentages for each of the five time points are logged and then the clustering procedure is used to group LSOAs which have similar trajectories. Figure 4 shows median log overcrowding by clusters determined using the kml package. The clusters themselves are mapped in Fig. 5. Most area types correspond to consistently decreasing overcrowding over the period 1971 to 2011. Clusters E and F (concentrated in parts of London and Birmingham, and some other urban areas) show high median levels of overcrowding at all time points. Cluster F (with most notable concentrations in parts of outer London) shows its lowest median levels of overcrowding in 1991, with a subsequent increase. Clusters A (dominating more rural areas across England) and C (found across England, with a particular concentration in the north east) start from relatively high levels and both show considerable decreases over the time period. The map shows clearly that similar overcrowding trajectories are found in different areas of the country, and the distinctions between areas with high or low population densities are obvious.
The trajectory classifications were linked to the Output Area Classification super groups (ONS, 2022) and the two classifications were cross-tabulated (Table 6). This enables a fuller assessment of trajectories for particular types of areas. Note that the OAC is for one time point (2011) only. There are clear trends for some OAC super groups. For example, over 80% of the OAC super group ‘London Cosmopolitan’ is in clusters E or F (39% and 42% respectively). Clusters E and F have high levels of overcrowding at all time points, with small increases in overcrowding between 2001 and 2011. Some 47% of ‘English and Welsh Countryside’ LSOAs are in cluster A – representing a decrease in overcrowding year-on-year from 1971 to 1991 and relative stability between 2001 and 2011. For ‘Mining Heritage and Manufacturing’, 33% are in cluster A and 29% are in cluster C (with higher levels of overcrowding than in cluster A, but the same general trend over time). In the case of ‘Prosperous England’, 38% are in cluster A while 30% are in cluster B – with the lowest levels of overcrowding at all time points and decreasing levels between 1971 and 1991, and relative consistency between 1991 and 2011.
Rooms per Person
The average number of rooms per person is central to the debate over housing space inequality (Tunstall, 2015) and whether there is a shortage of housing or just a poor distribution. Over the period 1911 to 2011, Tunstall notes that the population of England and Wales doubled while the number of rooms tripled, and the rate of ‘low absolute housing consumption (overcrowding)’ reduced from 49 to 4%. However, while inequality in rooms per person, as measured using the Gini coefficient, reduced from the 1920s to the 1980s, it grew again from the 1990s and by 2011 had reached levels not seen for at 50 years. Tunstall offers several possible explanations — increased income inequality, a reduction in social housing, a rise of one person households, and the development of larger homes. While housing stock continued to grow after the 1980s, the space which was available to the least well-housed ten percent of the population did not rise, resulting in an increased unevenness in housing space. Gleeson (2015a) argues that the trends identified by Tunstall can be explained in large part by a growing spatial mismatch between the distribution of population and housing stock in recent decades, as urban population growth outpaced housing supply. In this section we use neighbourhood level data to shed light on this debate.
The spatial distribution of the number of rooms per person broadly mirrors the map of overcrowding, and for this reason maps of rooms per person are not included here. The number of rooms per person was clearly fewer in LSOAs in urban areas than elsewhere. For the period 1971 to 2011, the number of rooms per person decreased, or increased by only a small amount, in many urban areas. The most obvious trend in the number of rooms per person between 2001 and 2011 is a decrease in London (corresponding to increased overcrowding, as illustrated in Fig. 3C) and the south east. While in most areas the Census year in which the number of rooms per person was smallest in each LSOA was 1971, there are notable exceptions, especially parts of the north of outer London where figures were smallest in 2011, plus a few areas where figures were smallest in 1981. A reduction in the number of rooms per person between 2001 and 2011 in outer London and, to a lesser degree, other urban areas is also observed by Dorling and Thomas (2016). Figure 6 shows change in rooms per person by initial density, 2001–2011. The plot for London indicates that the number of rooms per person fell in most areas while the figure increased in most areas of all other regions, although more densely populated areas saw a decrease in the number of rooms per person.
Households are, on average, larger in urban areas (particularly London, Birmingham, and Manchester) than elsewhere (Fig. 7A). Between 1971 and 2011, the average household size increased by more than one person in many parts of London (especially in parts of the east of London) and other urban areas including Birmingham, Bradford, and Oldham (Fig. 7B). This broad spatial pattern is also found between 2001 and 2011 specifically, as shown in Fig. 7C. Note that the maximum increase of 2.66 in Fig. 7C corresponds to a zero household count in 1971 (in this case, the household size was recorded as zero in 1971 and 2.66 in 2011); the maximum increase where there were households in 1971 and 2011 is 2.40. Average household sizes were at their largest in most LSOAs in 1971 or 1981, but in some parts of London and other urban areas average household sizes were largest in 2011. As was shown previously, overcrowding is largest in areas with large BAME populations, and the same is true of areas with large average household sizes. Catney and Simpson (2014) show that high average numbers of adults per household for South Asian groups persisted over the period 1991–2001. Over this time period economic circumstances improved for these groups, suggesting that preferences, rather than economic situations, drive household sizes. This is supported by the decline in overcrowding which was more marked for the Pakistani and Bangladeshi groups than for other groups. If there was an economic driver for large average household sizes, reduced overcrowding would be expected to be associated with a decline in average household sizes (in terms of adult sharing). Rather, reduced overcrowding in these groups is associated with a decline in fertility. In contrast, in London there is evidence that large household sizes are at least in part an involuntary consequence of the housing shortage. For example, Bentley and McCallum (2019) find that increases in household size and in young adults living with parents at regional level are linked to house price growth, suggesting a causal relationship between housing supply shortfalls and larger household sizes.
Figure 8 shows change in average household size by initial density, 1971–2011. Across England as a whole, the average household size fell between 1971 and 2011, but least of all in places with higher population densities in 1971; at regional levels the East and North East buck this trend, with similar declines in average household sizes throughout the density distribution.
Focusing on the latter half of this period and plotting the change in average household size between 1991 and 2011 by initial (1991) density (Fig. 9) shows some trends in common with change between 1971 and 2011 (Fig. 8). However, this time there is a positive relationship between initial density and change in average household size in every region, and most neighbourhoods in London see rising average household sizes.
Correlations Between Measures of Housing Space
The measures used in this study are distinct but inter-related: the degree of similarity can be measured using the Spearman’s rank correlation coefficient (indicated by rs). Values close to -1 or 1 suggest that the two measures contain the same information (they are strongly related), while values close to zero indicate that the measures contain distinct information. There is a strong relationship (with a correlation coefficient (rs) of -0.83) between overcrowding and rooms per person, and weaker correlations between overcrowding and average household size (rs = 0.21), and average household size and rooms per person (rs = 0.30). When the number of rooms per person in a household falls below 1 (or equivalently, persons per room rises above 1) it is considered overcrowded. An area with a highly unequal distribution of housing may therefore have an overcrowding problem even if its values on the other measures are not unusual.