GHS 2005 Appendix C: Sampling Errors - UK Data Servicedoc.ukdataservice.ac.uk/doc/6716/mrdoc/pdf/2005_sampling_errors.… · sampling frame, estimates may be biased because non-respondents

General Household Survey, 2005

Sampling Errors

General Household Survey 2005

Sampling Errors

Appendix C

Crown copyright

Office for National Statistics

Government Buildings

Cardiff Rd

Newport

NP10 8XG

Tel: 01633 812630

Email: [email protected]

Office for National Statistics: November 2006

UK Data Archive Study Number 6716 - General Lifestyle Survey, 2000-2008: Secure Data Service Access


Sampling Errors Tables in this appendix present estimates of sampling errors for some of the main

variables used in this report, taking into account the complex sample design of the survey.

Sources of error in surveys

Survey results are subject to various sources of error. The total error in a survey estimate is

the difference between the estimate derived from the sample data collected and the true

value for the population. The total error is made up of two main types: systematic and

random error.

Systematic error

Systematic error occurs when data are consistently biased in a certain way, such that the

variation from the true values for the population will not average to zero over repeats of

the survey. For example, if a certain section of the population is excluded from the

sampling frame, estimates may be biased because non-respondents to the survey have

different characteristics to respondents. Another cause of bias may be that interviewers

systematically influence responses in one way or another. Substantial efforts have been

made to avoid systematic errors, for example, through extensive interviewer training and

by weighting the data collected for non-response.

Random error

Random error, or bias, is the variation in sample data from the true values for the

population, which occurs by chance. This type of error is expected to average to zero over a

number of repeats of the survey. Random error may result from sources such as variation

in respondents’ interpretation of the survey questions, or interviewer variation. Efforts are

made to minimise these effects through pilot work and consistent interviewer training.

Sampling errors

An important component of random error is sampling error, which arises because the

variable estimates are based on a sample rather than a full census of the population. The

results obtained for any single sample would be likely to vary slightly from the true values

for the population. The difference between the estimates derived from the sample and the

true population values is referred to as

the sampling error. The amount of variation can generally be reduced by increasing the

size of the sample, and by improving the sample design. Sampling errors have been

measured for estimates derived from the General Household Survey (GHS), and these may

be used to assess the accuracy of the estimates presented in this report.

Office for National Statistics: November 2006 page 1


Sampling Errors

Calculating standard errors

Unlike non-sampling errors, it is possible to estimate the size of sampling error, by

calculating the standard error of the survey estimates. The standard error (se) of a

percentage p, based on a simple random sample of size n is calculated by the formula,

se(p)srs=√(p(100-p)/n).

The GHS uses a multi-stage sample design, which involves both clustering and

stratification (see Appendix B). The complexity of the design means that sampling errors

calculated on the basis of a simple random sample design will not reflect the true variance

in the survey estimates. Clustering can lead to a substantial increase in sampling error if

the households or individuals within the primary sampling units (PSUs) are relatively

homogenous but the PSUs differ from one another. By contrast, stratification tends to

reduce sampling error and is particularly effective when the stratification factor is related

to the characteristics of interest on the survey.

Because of the complexity of the GHS sample design, the size of the standard error

depends on how the characteristic of interest is spread within and between the PSUs and

strata. The method used to calculate the standard errors for the survey takes this into

account. It explicitly allows for the fact that the estimated values (percentages and means)

are ratios of two survey estimates: the number with the characteristic of interest is the

numerator (y) and the sample size is the denominator (x), both of which are subject to

random error.

The standard error of a survey estimate is found by calculating the positive square root of

the estimated variance of the ratio. The formula used to estimate the variance of a ratio

estimator r (where r = y/x) is shown below.

var(r) = 1 [var(y) + r2var (x) - 2r cov(y,x)]

x2

Var(r) is the estimate of the variance of the ratio, r, expressed in terms of var(y) and

var(x) which are the estimated variances of y and x, and cov(y,x) which is their estimated

covariance. The resulting estimate is only valid if the denominator (x) does not vary too

greatly. The method compares the differences between totals for adjacent PSUs (postal

sectors) in the characteristic of interest. The ordering of PSUs reflects the ranking of postal

sectors on the stratifiers used in the sample design.

Design factors

The design factor, or deft, of an estimate p is the ratio of the complex standard error of p to

the standard error of p that would have resulted had the survey design been a simple

random sample of the same size.

deft(p) = se(p)

sesrs(p)

This is often used to give a broad indication of the effect of the clustering on the reliability

of estimates. The size of the design factor varies between survey variables reflecting the



Sampling Errors degree to which a characteristic of interest is clustered within PSUs, or is distributed

between strata. For a single variable the size of the design factor also varies according to

the size of the subgroup on which the estimate is based, and on the distribution of that

subgroup between PSUs and strata. Design factors below 1.0 show that the complex

sample design improved on the estimate that we would have expected from a simple

random sample, probably due to the benefits of stratification. Design factors greater than

1.0 show less reliable estimates than might be gained from a simple random sample, due to

the effects of clustering. Design factors equal to 1.0 indicate no difference in the survey

design on the reliability of the estimate.

The formula to calculate the standard error of the difference between two percentages for a

complex sample design is:

se(p1-p2)=√[deft21(p1(100-p1)/n1)+ deft22 (p2(100-p2)/n2)].

where p1 and p2 are observed percentages for the two sub-samples and n1 and n2 are the

sub-sample sizes.

Confidence intervals

The estimate produced from a sample survey will rarely be identical to the population

value, but statistical theory allows us to measure its accuracy. A confidence interval can be

calculated around the estimated value, which gives a range in which the true value for the

population is likely to fall. The standard error measures the precision with which the

estimates from the sample approximate to the true population values and is used to

construct the confidence interval for each survey estimate.

The 95% confidence intervals have been calculated for each estimated value presented.

These are known as such, because if it were possible to repeat the survey under the same

conditions a number of times, we would expect 95% of the confidence intervals calculated

in this way to contain the true population value for that estimate. When assessing the

results of a single survey, it is usual to assume that there is only a 5% chance that the true

population value falls outside the 95% confidence interval calculated for each survey

estimate. To construct the bounds of the confidence interval, 1.96 times the standard error

is subtracted from, and added to, the estimated value, since under a normal distribution,

95% of values lie within 1.96 standard errors of the mean value. The confidence interval is

then given by:

p +/- 1.96 x se(p).

The 95% confidence interval for the difference between two percentages is given by:

(p1-p2) +/- 1.96 x se (p1-p2).

If this confidence interval includes zero then the observed difference is considered to be a

result of chance variation in the sample. If the interval does not include zero then it is

unlikely (less than 5% probability) that the observed difference could have occurred by

chance.



Sampling Errors

Standard errors for the 2005 GHS

The standard errors were calculated on weighted data using STATA1. Weighting for

different sampling probabilities results in larger sampling errors than for an equal-

probability sample without weights. However, weighting which uses population totals to

control for differential non-response tends to lead to a reduction in the errors. The method

used to calculate the sampling errors correctly allows for the inflation in the sampling

errors caused by the first type of weighting but, in treating the second type of weighting in

the same way as the first, incorrectly inflates the estimates further. Therefore the standard

errors and defts presented are likely to be slight over-estimates. Weighted data were used

so that the values of the percentages and means were the same as those in the substantive

chapters of the report.

Tables C.1 to C.12 show the standard error, the 95% confidence interval and the deft for

selected survey estimates. The tables do not cover all the topics discussed in the report but

show a selection of estimates.

For the design factors of household based estimates, one was below 1, one fifth (20%) were

below 1.1, just over two fifths (43%) were less than 1.2, and just over two-thirds (70%) of

the defts were less than 1.3. There were three cases (10% of all the household-based

estimates) where the deft was 1.5 or greater. The higher defts were mostly for tenure and

accommodation type (Table C.1) where the effects of clustering lead to a loss of precision

compared with that of a simple random sample. The defts that were below 1.1 were in part

for the number of persons in the household and household type, indicating that

stratification has increased the precision of the sample over a simple random sample for

these estimates of household variables.

For the design factors of person based estimates, 2% were below 1, about a tenth (12%)

were below 1.1, nearly two-thirds (63%) were less than 1.2, and half (50%) of the defts were

less than 1.3. Three per cent of the defts were 1.5 or greater, including many of those for

estimates of ethnicity, shown in Table C.6. As well as clustering in the same sectors, people

from the same ethnic backgrounds will generally cluster within the same households, and

so estimates have high sampling errors and high defts. In contrast, estimates broken down

by gender will generally have lower sampling errors because there is often one man and

one woman in a household; for example, the estimates of males and females in the

population have defts of 0.8 (Table C.4).

Estimating standard errors for other survey measures

The standard errors of survey measures, which are not presented in the tables and for

sample subgroups may be estimated by applying an appropriate value of deft to the

sampling error. The choice of an appropriate value of deft will vary according to whether

the basic survey measure is included in the tables. Since most deft values are relatively

small (1.3 or less) the absolute effect of adjusting sampling errors to take account of the

survey’s complex design will be small. In most cases it will result in an increase of less than

30% over the standard error assuming a simple random sample. Whether it is considered



Sampling Errors necessary to use deft or to use the basic estimates of standard errors assuming a simple

random sample is a matter of judgement and depends chiefly on the use to which the

survey results are to be put.

Notes and references

1. STATA is a statistical analysis software package. For further details of the method of

calculation see: Elliot D. A comparison of software for producing sampling errors on social

surveys. SSD Survey Methodology Bulletin 1999; 44: 27-36.



Sampling Errors

Table C1 Standard errors and 95% confidence intervals for household tenure, household type and accommodation type

Base Characteristic % (p ) Deft

All householdsHousehold type1 adult aged 16-59 15.9 12802 0.38 15.1 - 16.6 1.22 adults aged 16-59 16.4 12802 0.36 15.7 - 17.1 1.1Youngest person aged 0-4 10.6 12802 0.27 10.1 - 11.1 1.0Youngest person aged 5-15 15.6 12802 0.34 14.9 - 16.2 1.13 or more adults 10.2 12802 0.34 9.5 - 10.9 1.32 adults, 1 or both aged 60 or over 15.9 12802 0.36 15.2 - 16.6 1.11 adult aged 60 or over 15.4 12802 0.32 14.8 - 16.1 1.0

TenureOwner occupied, owned outright 30.3 12800 0.47 29.4 - 31.2 1.2Owner occupied, with mortgage 38.4 12800 0.51 37.4 - 39.4 1.2Rented from council 12.3 12800 0.46 11.4 - 13.2 1.6Rented from housing association 7.5 12800 0.31 6.8 - 8.1 1.3Rented privately, unfurnished 8.2 12800 0.29 7.7 - 8.8 1.2Rented privately, furnished 3.3 12800 0.26 2.8 - 3.8 1.6

Accommodation typeDetached house 21.9 12802 0.48 21.0 - 22.9 1.3Semi-detached house 30.8 12802 0.55 29.8 - 31.9 1.3Terraced house 27.7 12802 0.58 26.6 - 28.8 1.5Purpose-built flat or maisonette 15.7 12802 0.48 14.8 - 16.7 1.5Converted flat or maisonette/rooms 3.8 12802 0.27 3.2 - 4.3 1.6With business premises/other 0.1 12802 0.02 0.0 - 0.1 0.9

Unweighted sample size

Standard error of p

95% confidence intervals



Sampling Errors

Table C2 Standard errors and 95% confidence intervals for number of persons and cars at each household


All householdsNumber of persons1 31.3 12802 0.48 30.4 - 32.3 1.22 35.0 12802 0.46 34.1 - 35.9 1.13 14.8 12802 0.33 14.2 - 15.5 1.14 12.9 12802 0.30 12.3 - 13.5 1.05 4.2 12802 0.17 3.9 - 4.6 1.06 or more 1.7 12802 0.13 1.5 - 2.0 1.1

Number of cars/light vans1 44.8 12802 0.46 43.9 - 45.7 1.02 or more 30.1 12802 0.44 29.2 - 30.9 1.1none 25.1 12802 0.45 24.2 - 26.0 1.2



Standard error of p

Table C3 Standard errors and 95% confidence intervals for households' ownership of selected consumer durables


All householdsSelected consumer durables

Home computer 63.4 12802 0.50 62.4 - 64.3 1.2

Washing machine 36.6 12802 0.50 35.7 - 37.6 1.2



Standard error of p



Sampling Errors

Table C4 Standard errors and 95% confidence intervals for age and sex


SexAll persons Male 48.7 30069 0.24 48.2 - 49.1 0.8

Female 51.3 30069 0.24 50.9 - 51.8 0.8

AgeAll persons 0-4 5.7 30069 0.15 5.4 - 6.0 1.1

5-15 13.8 30069 0.25 13.3 - 14.3 1.316-44 39.8 30069 0.40 39.0 - 40.6 1.445-64 25.0 30069 0.33 24.4 - 25.6 1.365-74 8.5 30069 0.20 8.1 - 8.9 1.275 and over 7.1 30069 0.19 6.8 - 7.5 1.3

All males 0-4 6.0 14580 0.21 5.6 - 6.4 1.15-15 14.3 14580 0.34 13.7 - 15.0 1.216-44 40.3 14580 0.53 39.2 - 41.3 1.345-64 25.3 14580 0.41 24.5 - 26.1 1.165-74 8.3 14580 0.25 7.8 - 8.8 1.175 and over 5.8 14580 0.19 5.4 - 6.2 1.0

All females 0-4 5.4 15489 0.18 5.0 - 5.7 1.05-15 13.3 15489 0.30 12.8 - 13.9 1.116-44 39.4 15489 0.42 38.6 - 40.2 1.145-64 24.7 15489 0.37 24.0 - 25.4 1.165-74 8.7 15489 0.22 8.3 - 9.2 1.075 and over 8.4 15489 0.25 7.9 - 8.9 1.1



Standard error of p



Sampling Errors

C5 Standard errors and 95% confidence intervals for marital status


Marital StatusAll persons aged 16 and over Married 51.0 23750 0.50 50.0 - 51.9 1.5

Cohabiting 9.9 23750 0.31 9.2 - 10.5 1.6Single 23.5 23750 0.47 22.6 - 24.5 1.7Widowed 7.1 23750 0.18 6.8 - 7.5 1.1Divorced 6.1 23750 0.17 5.8 - 6.4 1.1Separated 2.1 23750 0.11 1.9 - 2.4 1.2

Men aged 16 and over Married 52.9 11367 0.61 51.7 - 54.1 1.3Cohabiting 10.2 11367 0.32 9.6 - 10.9 1.1Single 26.7 11367 0.60 25.5 - 27.9 1.5Widowed 3.2 11367 0.18 2.9 - 3.6 1.1Divorced 4.8 11367 0.17 4.5 - 5.1 0.9Separated 1.8 11367 0.15 1.5 - 2.0 1.2

Women aged 16 and over Married 49.2 12383 0.52 48.1 - 50.2 1.2Cohabiting 9.5 12383 0.30 8.9 - 10.1 1.1Single 20.6 12383 0.52 19.6 - 21.6 1.4Widowed 10.7 12383 0.30 10.1 - 11.3 1.1Divorced 7.3 12383 0.26 6.8 - 7.8 1.1Separated 2.5 12383 0.15 2.2 - 2.8 1.1

All persons aged 16 to 24 Married 3.8 2990 0.45 2.9 - 4.7 1.3Cohabiting 11.4 2990 0.82 9.8 - 13.0 1.4Single 83.9 2990 0.99 82.0 - 85.9 1.5Widowed 0.0 2990 0.03 0.0 - 0.1 1.0Divorced 0.2 2990 0.09 0.0 - 0.4 1.1Separated 0.4 2990 0.12 0.2 - 0.7 1.0






All persons aged 75 and over Married 42.8 2126 1.27 40.3 - 45.3 1.2Cohabiting 1.0 2126 0.29 0.5 - 1.6 1.3Single 6.2 2126 0.60 5.0 - 7.3 1.2Widowed 46.2 2126 1.32 43.6 - 48.8 1.2Divorced 3.0 2126 0.39 2.2 - 3.7 1.1Separated 0.8 2126 0.19 0.4 - 1.2 1.0


Standard error of p




Sampling Errors

Table C.6 Standard errors and 95% confidence intervals for ethnic origin*

Base1 Characteristic % (p ) Deft

All persons aged 16 and over Ethnic Origin

White 91.0 23706 0.46 90.1 - 91.9 2.5Mixed Race 0.8 23706 0.08 0.6 - 1.0 1.4Asian-Indian 2.1 23706 0.21 1.6 - 2.5 2.3Asian- Pakistani, Bangladeshi, Other 2.4 23706 0.25 2.0 - 2.9 2.5Black Caribbean 1.1 23706 0.12 0.9 - 1.4 1.8Black African 1.3 23706 0.12 1.0 - 1.5 1.7Other 1.4 23706 0.14 1.1 - 1.6 1.9

* Other includes other Black groups. Information on those giving no answer has not been presented1 These estimates are based on 2005 data only, wheras in the report estimates are based on 2001, 2002 and 2003 data combined.We would expect the defts to be very similar for the combined years estimates.


Standard error of p


Table C.7 Standard errors and 95% confidence intervals for education level


Education levelAll persons aged 16 to 69 Higher education 28.5 20174 0.49 27.6 - 29.5 1.5

Other qualifications 45.2 20174 0.44 44.4 - 46.1 1.3None 26.2 20174 0.46 25.3 - 27.1 1.5

All men aged 16 to 69 Higher education 29.4 9336 0.59 28.2 - 30.5 1.3Other qualifications 46.8 9336 0.59 45.6 - 47.9 1.1None 23.9 9336 0.55 22.8 - 24.9 1.2

All women aged 16 to 69 Higher education 27.8 10838 0.56 26.7 - 28.9 1.3Other qualifications 43.9 10838 0.53 42.8 - 44.9 1.1None 28.3 10838 0.53 27.3 - 29.3 1.2



Standard error of p



Sampling Errors

Table C8 Standard errors and 95% confidence intervals for socio-economic classification and employment status of adults


Socio-economic classification

All persons aged 16 and over Higher managerial and professional 11.0 23750 0.26 10.5 - 11.5 1.3

Lower managerial and professional 22.8 23750 0.35 22.1 - 23.5 1.3

Intermediate 12.9 23750 0.24 12.4 - 13.4 1.1

Small employers and own account 7.8 23750 0.19 7.4 - 8.2 1.1

Lower supervisory and technical 9.0 23750 0.22 8.6 - 9.5 1.2

Semi-routine 17.6 23750 0.31 16.9 - 18.2 1.3

Routine 14.1 23750 0.30 13.5 - 14.7 1.3

Never worked and long-term unemployed 4.9 23750 0.19 4.5 - 5.2 1.4

Higher managerial and professional 16.3 10525 0.42 15.5 - 17.1 1.2

Lower managerial and professional 21.7 10525 0.42 20.9 - 22.5 1.1Intermediate 6.3 10525 0.26 5.8 - 6.8 1.1Small employers and own account 11.2 10525 0.31 10.6 - 11.8 1.0

Lower supervisory and technical 13.3 10525 0.38 12.5 - 14.0 1.2Semi-routine 12.2 10525 0.36 11.5 - 12.9 1.1Routine 15.7 10525 0.42 14.9 - 16.5 1.2Never worked and long-term unemployed 3.4 10525 0.23 2.9 - 3.8 1.3

Higher managerial and professional 6.1 11546 0.24 5.7 - 6.6 1.1Lower managerial and professional 23.9 11546 0.45 23.0 - 24.7 1.1Intermediate 18.9 11546 0.36 18.2 - 19.7 1.0

Small employers and own account 4.6 11546 0.21 4.2 - 5.0 1.1Lower supervisory and technical 5.1 11546 0.22 4.7 - 5.6 1.1Semi-routine 22.5 11546 0.44 21.6 - 23.3 1.1Routine 12.7 11546 0.34 12.0 - 13.3 1.1

Never worked and long-term unemployed 6.2 11546 0.27 5.7 - 6.7 1.2

Higher managerial and professional 11.8 9778 0.38 11.1 - 12.5 1.2Lower managerial and professional 24.2 9778 0.49 23.3 - 25.2 1.1Intermediate 13.6 9778 0.36 12.9 - 14.3 1.0Small employers and own account 6.3 9778 0.27 5.7 - 6.8 1.1Lower supervisory and technical 8.1 9778 0.30 7.5 - 8.7 1.1

Semi-routine 16.9 9778 0.43 16.1 - 17.8 1.1Routine 12.0 9778 0.40 11.2 - 12.8 1.2Never worked and long-term unemployed 7.1 9778 0.31 6.5 - 7.7 1.2


Lower managerial and professional 23.8 7465 0.54 22.8 - 24.9 1.1Intermediate 11.9 7465 0.37 11.1 - 12.6 1.0Small employers and own account 9.8 7465 0.38 9.0 - 10.5 1.1Lower supervisory and technical 8.8 7465 0.34 8.2 - 9.5 1.0Semi-routine 17.4 7465 0.49 16.4 - 18.3 1.1Routine 13.8 7465 0.46 12.9 - 14.7 1.2Never worked and long-term unemployed 2.4 7465 0.20 2.0 - 2.8 1.1

Higher managerial and professional 7.9 2731 0.57 6.8 - 9.0 1.1Lower managerial and professional 18.6 2731 0.86 16.9 - 20.3 1.2

Intermediate 11.8 2731 0.63 10.6 - 13.1 1.0Small employers and own account 8.9 2731 0.59 7.7 - 10.1 1.1Lower supervisory and technical 11.9 2731 0.69 10.6 - 13.3 1.1Semi-routine 17.9 2731 0.78 16.4 - 19.4 1.1Routine 19.7 2731 0.84 18.1 - 21.4 1.1Never worked and long-term unemployed 3.3 2731 0.41 2.5 - 4.1 1.2


Lower managerial and professional 17.5 2097 0.85 15.8 - 19.1 1.0Intermediate 14.1 2097 0.84 12.5 - 15.8 1.1Small employers and own account 6.9 2097 0.62 5.7 - 8.1 1.1

Lower supervisory and technical 10.8 2097 0.66 9.5 - 12.1 1.0Semi-routine 20.8 2097 0.96 18.9 - 22.7 1.1Routine 18.8 2097 0.96 16.9 - 20.6 1.1Never worked and long-term unemployed 4.4 2097 0.46 3.5 - 5.3 1.0

Employment statusAll persons aged 16 & over In employment 59.5 23381 0.46 58.6 - 60.4 1.4

Unemployed 2.4 23381 0.12 2.2 - 2.7 1.2Economically inactive 38.1 23381 0.46 37.2 - 39.0 1.5

In employment 65.9 11135 0.57 64.8 - 67.0 1.3Unemployed 3.0 11135 0.18 2.7 - 3.4 1.1Economically inactive 31.1 11135 0.55 30.0 - 32.2 1.3

In employment 53.7 12246 0.53 52.7 - 54.7 1.2

Unemployed 1.9 12246 0.13 1.6 - 2.1 1.1Economically inactive 44.5 12246 0.53 43.4 - 45.5 1.2

All persons aged 65 to 74

All persons aged 75 and over

All men aged 16 and over

All women aged 16 and over






Standard error of p




Sampling Errors

Table C.9 Standard errors and 95% confidence intervals for health measures


Self-reported sicknessAll persons Longstanding illness 32.8 28206 0.46 31.9 - 33.7 1.7

Limiting longstanding illness 18.2 29708 0.30 17.6 - 18.8 1.3Restricted activity in the last 14 days 13.1 28210 0.29 12.5 - 13.7 1.4

All males Longstanding illness 32.3 13394 0.51 31.3 - 33.3 1.3Limiting longstanding illness 17.0 14353 0.36 16.2 - 17.7 1.2Restricted activity in the last 14 days 11.4 13397 0.34 10.8 - 12.1 1.2

All females Longstanding illness 33.3 14812 0.54 32.3 - 34.4 1.4Limiting longstanding illness 19.3 15355 0.40 18.5 - 20.1 1.3Restricted activity in the last 14 days 14.6 14813 0.37 13.9 - 15.3 1.3

All persons aged Longstanding illness 12.0 1819 0.86 10.3 - 13.7 1.10 to 4 Limiting longstanding illness 3.8 1901 0.46 2.9 - 4.7 1.1

Restricted activity in the last 14 days 9.1 1892 0.75 7.6 - 10.6 1.1









Longstanding illness 64.4 2047 1.25 61.9 - 66.8 1.2Limiting longstanding illness 45.2 2115 1.17 42.9 - 47.5 1.1Restricted activity in the last 14 days 21.7 2046 1.11 19.5 - 23.9 1.2


Standard error of p


All persons aged 75+



Sampling Errors

Table C.10 Standard errors and 95% confidence intervals for cigarette smoking


Cigarette smokingAll persons aged 16 & over Current cigarette smoker 23.9 21665 0.37 23.2 - 24.6 1.3

Ex-regular cigarette smoker 23.7 21665 0.35 23.0 - 24.4 1.2Never regularly smoked cigarettes 52.5 21665 0.46 515.6 - 53.4 1.4

Current cigarette smoker 25.3 10038 0.50 24.3 - 26.3 1.2Ex-regular cigarette smoker 27.3 10038 0.52 26.2 - 28.3 1.2Never regularly smoked cigarettes 47.5 10038 0.61 46.3 - 48.7 1.2

Current cigarette smoker 22.7 11627 0.42 21.9 - 23.5 1.1Ex-regular cigarette smoker 20.5 11627 0.39 19.8 - 21.3 1.0Never regularly smoked cigarettes 56.8 11627 0.53 55.8 - 57.8 1.2



Standard error of p





Sampling Errors

Table C11 Standard errors and 95% confidence intervals for alcohol consumption (maximum daily amount)


Alcohol consumption(maximum daily amount)

All men aged 16 and over Drank nothing last week 27.8 10028 0.54 26.7 - 28.8 1.2Drank up to 4 units 37.6 10028 0.56 36.5 - 38.7 1.2Drank more than 4 and up to 8 units 16.0 10028 0.39 15.3 - 16.8 1.1Drank more than 8 units 18.6 10028 0.45 17.7 - 19.5 1.2

All women aged 16 and over Drank nothing last week 43.1 11617 0.57 42.0 - 44.2 1.2Drank up to 3 units 36.8 11617 0.49 35.8 - 37.7 1.1Drank more than 3 and up to 6 units 12.3 11617 0.34 11.6 - 13.0 1.1Drank more than 6 units 7.8 11617 0.31 7.2 - 8.5 1.2

All aged 16 to 24 Drank nothing last week 40.4 2384 1.36 37.7 - 43.1 1.4Drank up to 4/3 units 20.5 2384 0.98 18.6 - 22.4 1.2Drank more than 4/3 and up to 8/6 units 13.6 2384 0.80 12.1 - 15.2 1.1Drank more than 8/6 units 25.5 2384 1.07 23.4 - 27.6 1.2



All aged 65 and over Drank nothing last week 46.5 4692 0.84 44.9 - 48.2 1.2Drank up to 4/3 units 44.3 4692 0.80 42.7 - 45.9 1.1Drank more than 4/3 and up to 8/6 units 7.0 4692 0.41 6.2 - 7.8 1.1Drank more than 8/6 units 2.2 4692 0.23 1.7 - 2.6 1.1

All aged 16 and over Drank nothing last week 35.9 21645 0.45 35.1 - 36.8 1.4Drank up to 4/3 units 37.2 21645 0.41 36.4 - 38.0 1.3Drank more than 4/3 and up to 8/6 units 14.0 21645 0.27 13.5 - 14.6 1.1Drank more than 8/6 units 12.9 21645 0.30 12.3 - 13.4 1.3


Standard error of p




Sampling Errors


Table C12 Standard errors and 95% confidence intervals for number of cohabitations


Number of cohabitationsAll women aged 16 to 59 None 83.1 8065 0.49 82.2 - 84.1 1.2

One 13.0 8065 0.42 12.2 - 13.8 1.1Two or more 3.9 8065 0.22 3.4 - 4.3 1.0

All men aged 16 to 59 None 84.4 6975 0.49 83.5 - 85.4 1.1One 10.6 6975 0.40 9.8 - 11.4 1.1Two or more 5.0 6975 0.29 4.4 - 5.5 1.1

All people aged 16 to 24 None 91.3 2369 0.60 90.1 - 92.5 1.0One 7.4 2369 0.50 6.4 - 8.4 0.9Two or more 1.3 2369 0.25 0.8 - 1.8 1.1







Standard error of p


GHS 2005 Appendix C: Sampling Errors - UK Data Servicedoc.ukdataservice.ac.uk/doc/6716/mrdoc/pdf/2005_sampling_errors.… · sampling frame, estimates may be biased because non-respondents

Documents