Hcond - condition survey software

• Contents
• Contact us

6 Sample Size

The sample size needs careful thought before the survey commences. If the stock is varied in its nature or if the stock is known to be very different in terms of condition a large sample may be required- possibly up to 50% or so. Similarly, if the survey aims to collect detailed data on catch-up and short term repairs (including specification details) a large sample will also be required. However, if the stock is of a similar nature and the survey aim is to collect long-term projections a small sample will suffice. Where both the above apply (ie some old units and some units on large estates) the survey sample should be stratified.

Example of stratified sample

Statistical Reliability

In practice the only way of determining the margin of error in a sample is to compare it against the whole stock. This is obviously a counter productive exercise. However, where the stock is relatively similar sample of 5%-10% may be acceptable. Consider the table below. It's a bit complex at first sight but does show the accuracy of a number of samples. 

We surveyed over 150 houses for an association in 1996. About 40% of the dwellings were old acquired houses in various states of repair. The remaining houses were built in the 1970s and 1980s. The association required data on catch-up repairs and projections over the next 10 years. The average cost of the repairs/renewals for each dwelling (at 1996 prices) was £8000 - shown as the horizontal black line in the chart. We then fed all the data (average costs per unit) into Excel and sampled it five times at various percentages, 1%, 2%, 5%, 10%, 20% 50% & 80%. For each sample size we asked Excel to randomly select the correct number of properties. The results are shown below.

At 1% the five random samples are all a long way short of the average. The lowest was about £300, the highest £5,300. This is because the 2 units selected each time were not representative of the whole.

At 2% the survey is more accurate. The four dwellings selected in each of the five attempts are quite consistent - although still below the average for the whole stock.

At 10% the average of the 5 random selections is just over £8,000. More importantly none of the five selections is far off the overall average of £8,000.

Over 10%, the sample does not make much difference.

This is, perhaps, a fairly crude example, but it does illustrate the nature of sampling. There is a paper on sampling in the download section which takes a more mathematical approach. What is worth remembering is that the number of dwellings which should be surveyed is related to the standard deviation (ie the spread of costs) - not the number of dwellings.