COVER SHEET

Skitmore, Martin R. and Wilcox, John (1994) Estimating processes of

smaller builders. Construction Management and Economics
12(2):pp. 139-154.

Copyright 1994 Taylor & Francis.

Accessed from: http://eprints.qut.edu.au/archive/00004438/
 ESTIMATING PROCESSES OF
SMALLER BUILDERS

Paper prepared for

Construction Management and Economics

Martin Skitmore
Department of Surveying
University of Salford
Salford M5 4WT

John Wilcock
Department of Construction and Surveying
The Fylde College of Technology and Arts
Bispham
Blackpool
Lancashire

June 1993 (Revised)

ESTIMATING PROCESSES OF

SMALLER BUILDERS

ABSTRACT

The paper describes a study of the way smaller builders price

bills of quantities items for competitive tender. A series of

interviews revealed some marked differences between normal

practise and literature-based prescriptions. An experiment was

conducted in which eight practising builders' estimators were

separately presented with a representative sample of 36 bill of

quantities items taken from groundwork, in-situ concrete work

and masonry sections. The estimators stated the method they

would normally use to price each item, their 'normal' price rate

and their highest/lowest price rate. The results showed that

only half the items would be priced by the prescribed 'detailed'

method, the remainder being priced mainly by 'experience'.

Analysis by work section, item rate, item quantity, item total,

item labour content, contribution to the total of the bill, the

standard deviation of the inter-estimator intra-item rates and

totals, and their coefficients of variation, skewness and

kurtosis indicated that the item total was the main factor

determining the rating method used, although this varied in

importance between work sections. An intra-estimator intra-item

analysis of pricing variability generally confirmed the

assumption of a constant coefficient of variation.

Keywords: Pricing, bills of quantities, item rates, variability.

 2

INTRODUCTION

All procurement systems require, at some stage, a price to be

agreed between the procurer and constructor. For most building

projects, this price is tendered by a builder before carrying

out the work. In this case the tender price is based on

documents specifying the end product in the form of a bill of

quantities, specifications and drawings, drawings only or just

the procurer's brief.

Although there is a wealth of prescriptive literature,

surprisingly little descriptive material is available concerning

the processes employed by builders in determining a tender

price. It is known however that all builders have estimating

personnel, or at least someone who compiles 'estimates' which

form the basis of tender prices. As a result, and the

prescriptive literature is in unanimous agreement, it is

generally assumed that tender prices are based on an estimate of

the likely costs of construction. Over the last 20 years

however, the assumption that tender prices are based on

builders' genuine estimates of future expenditure has been

increasingly questioned. Fine (1974), for example, has

introduced the notion of the "socially acceptable price",

implying that tender prices are based on the characteristics of

the finished product rather than the processes involved in

producing that product.

 3

Defining the actual concepts underlying the estimating process

is clearly a matter for empirical study. Empirical work in the

field is however fraught with difficulty. As 'estimators' are

familiar with the prescriptive literature, they tend to

rationalise their work in the terms of this literature.

Inevitably, they respond affirmatively to direct questions such

as "do you try to estimate likely production costs?" or "do you

take productivity into account?". Delving a little deeper

however reveals some paradoxes. The use of feedback is a prime

example. In one study of six builders' estimators (Hampson,

1979), it was found that only one estimator kept formal records

of site performance. If estimators are genuinely trying to

forecast actual costs, why do many not keep records of actual

costs? In the absence of a resolution to this paradox, the

argument exemplified by Fine must be considered to stand

unrefuted.

In the research reported in this paper, the procedures used by

builders in 'estimating' building work were examined. A series

of interviews is described in which nine smaller builders'

estimators revealed several marked differences between normal

practise and literature based prescriptions. One of these

differences was claimed to be in the method of 'pricing' items

in bills of quantities. This is the process, formal or

otherwise, of attaching a 'rate' to an item that is subsequently

multiplied by the 'quantity' to give a total 'price' for that

item (the procedure is termed 'rating' the item in this paper).

 4

To investigate this further, an experiment was conducted in

which eight practising builders' estimators were separately

presented with a representative sample of 36 bill of quantities

items taken from the groundwork, in-situ concrete work and

masonry sections. The estimators stated the method they would

normally use to rate each item, their 'normal' rate and their

highest/lowest rate. The results showed that only half the

items would be rated by the prescribed 'detailed' method, the

remainder being rated mainly by 'experience'. Analysis by work

section, item rate, item quantity, item total, item labour

content, contribution to the total of the bill, the standard

deviation of the inter-estimator intra-item (represented by the

term RA) rates and totals, and RA coefficients of variation,

skewness and kurtosis indicated that the item total was the main

factor associated with the rating method used, although this

varied in importance between work sections. Finally, the intra-

estimator intra-item (represented by the term AA) variability of

rates was examined in terms of the acknowledged distribution of

rates by estimators. In the realisation that this work has some

relevance in the formulation of measurement methods, Barnes'

(1971) assumption, ie., that "the estimation of unit costs of

the items is assumed to have a constant coefficient of variance"

was also tested.

THE SAMPLE

To evaluate current estimating practise, a separate interview

 5

and questionnaire survey was undertaken with a sample of average

practising builders' estimators. In this study, 'average'

estimators were taken to be those employed in typical building

companies. These were limited according to the resources at our

disposal and selected on the following considerations.

Size of company

It is difficult to define the size of building firms as small,

medium or large. Whilst Housing and Construction Statistics

present information relating to the number of employees, this is

not a completely satisfactory method of defining the size of a

firm. Most builders now employ labour-only or sub-contractors

to carry out the work while they undertake the management

functions and provide materials and site set-up. A more

accurate definition of size might be based on annual turnover,

but obtaining accurate data presents a problem, as many smaller

firms do not have to make such details public.

Most studies in the building industry centre on large

contracting organisations. However, such organisations are not

typical in the building industry. In 1987 the number of people

employed by private builders totalled 961,900. 67.92% of these

were employed by firms employing 114 people or less (Housing and

Construction Statistics 1977-87, Table 3.10) compared with 61%

in the previous decade (Housing and Construction Statistics,

1977-87, Table 3.14), and most building orders were for

 6

contracts of less than £2 million.

In view of this, a sample of smaller builders was selected,

'smaller' being defined as those companies employing 114 people

or less, as representing the average building company in the

industry.

Conditions of obtaining work

Current trade journals usually contain articles dealing with

high value prestige projects. These large projects are regarded

as prestigious not only by the client but by the architects and

others. They give the appearance of being relatively well

prepared and thought out with potential contractors being

presented with full tender documentation and allowed a

reasonable time to prepare their tenders. A considerable amount

of work however, all of a minor nature, is obtained and

undertaken by smaller firms. This then raises the question

concerning the conditions under which these smaller firms obtain

their work, i.e., are minor projects presented to builders in an

identical way to major projects?

There is little press coverage of projects typified by a £50,000

alteration to the local sewage works. This type of project,

when linked to a demanding client, limited design detailing, and

a small budget, often proves to be most demanding. Though it

does not necessarily follow that large builders do not undertake

 7

small contracts, it is certainly true that small builders

undertake only small contracts and therefore their experience is

restricted to and dominated by these kinds of conditions. It is

to be expected therefore, that this will be reflected in the

procedures of smaller builders.

Location

The Blackpool and Preston areas have a combined population of

526,458 (Key Statistics for Urban Areas, 1981), excellent

transportation links, and a light industry and service industry

base and a large tourist industry. This mixture of light

industry, service industry and a tourist related industry

combined with local and central government developments provides

a local economic climate suited to smaller builders. There are

few multi million pound projects and many projects are within

the capabilities of these builders. As a result, this area was

considered suitable to conduct research with smaller builders as

it has a suitable base from which they can operate (besides

being convenient for the researchers).

DATA COLLECTION AND ANALYSIS

Survey of builders

Ten smaller builders were selected at random from a

 8

comprehensive list of local builders in the Blackpool and

Preston area. These builders were contacted and one estimator

from each of nine of the firms agreed to take part in the study.

The nine estimators were interviewed separately and informally

in May and June 1990 and eight of these later completed a postal

questionnaire. All the estimators worked for general building

contractors who undertook all types of building work. Two firms

undertook some additional speculative housing, and none

undertook civil engineering work. The estimators interviewed

had an average of eighteen years experience and carried out the

estimating process regularly and personally. The interviews all

took place in the estimators' own offices and lasted for

approximately one hour, during which time freehand notes were

taken by the researchers.

The questions in both the interviews and questionnaires were

designed to be easily answered by practising builders'

estimators. All the estimators appeared to be interested in the

research and freely discussed all aspects of the estimating

process with enthusiasm.

The Interviews

The interviews were used as a contact point with the estimators

and to obtain their opinions on the differences between the

prescriptive literature and the realities of tender preparation.

The main issues discussed are summarised below.

 9

Decision to tender. All the estimators' firms tendered for all

the work sent to them at the time of the study. The amount of

detail and effort that went into tender preparation was

dependent on how much the firm wanted the work, a point

emphasised later in the data collected by the questionnaire.

Project appreciation. Attitudes to this varied. Drawings were

considered essential by all the estimators but varied in the way

they were studied. Some estimators studied the drawings prior

to rating, allowing for the complexity of the work in the item

rates. Others studied the drawings after rating the items,

adjusting for complexity by a lump sum in the Preliminaries.

None of the estimators interviewed undertook anything but the

briefest of pre-tender planning, a task usually carried out by

the estimators themselves. The normal procedure was to decide

whether or not it was feasible to complete the work in the time

specified in the tender documents. One estimator said that if

it was not considered feasible, he would then make an allowance

for overtime working by multiplying the liquidated damages per

week by the difference between the stated contract period and a

period the estimator considered reasonable and practicable.

Site visits prior to rating were not considered by all to be

essential and the practise varied from visiting one in ten sites

to visiting every site. Visiting the site also had different

meanings to different estimators. One estimator who tried to

 10

visit every site said that if he was tendering for, say, a

million pound project, he would drive past the site on his way

home. He would have a look to see if there was anything of

obvious interest.

This rather subjective approach of pre-tender planning and risk

appraisal was justified by the estimators on two grounds: (1)

the size and complexity of the projects under consideration were

within their recent experience; and (2) limiting the expense

involved in tender preparation was a valid way of minimising the

firms' opportunity costs.

Item rates. When the estimators rated items in detail they did

so in the manner prescribed in the literature ie., by careful

'build-up' of labour, material and plant costs from first

principles, use of supplier and subcontractor quotations, etc.

However such detailed analyses were only undertaken when an item

was considered to be of significant value. Items of small

financial importance were either ignored or rated at "what the

job would stand".

Another aspect rarely mentioned in the literature was that of

substituting alternative construction methods, particularly in

temporary works, eg the use of blockwork, in lieu of plywood, in

formwork to ground beams and foundations. Here the estimator

would allow in his rate for blockwork instead of the more

traditional plywood, if this provided a cheaper alternative,

even though plywood was specified in the tender documents.

 11

Completing the tender. Items were usually rated inclusive of

profit and overheads. All the estimators considered that rating

the attendance on nominated sub-contractors could only be done

with experience and luck. All the estimators said that the

special attendance items in bills of quantities were of little

use as a guide to rating. The estimators were also in agreement

on the methods used in rating the preliminaries section. All

the estimators rated only what they felt the Architect would

want on site, and not what was specified in the bills of

quantities - another approach not prescribed in the literature.

Generally. The interviews highlighted four main areas of

conflict between actual and prescribed tender pricing practices:

1) The limitation of estimating effort and tender preparation

costs by using experience to rate certain items rather than

detailed rate analysis. This method of rating items was

undertaken to some extent by the firms of the estimators

interviewed and appeared to be applied to items considered by

the estimator to be of insignificant value. This confirms

the view of Eastham (1990) who found some similar general

variations in pricing methods between contractors.

2) The rating of items inclusive of profit and overheads - such

adjudication practices that did take place being to 'fine

tune' the profit allowance.

 12

3) The limitation tender preparation costs by reducing

pre-tender planning and risk appraisal to an absolute

minimum.

4) The rating of preliminary items on a basis of what might

eventually be wanted rather than what was actually specified.

The Questionnaire

As a consequence of result 1) above, a questionnaire survey was

undertaken to gain further information on the methods of rating

individual items in bills of quantities. The questionnaire

contained typical bills of quantities items for Groundwork,

In-situ Concrete and Masonry (Appendix A). Estimators were

requested to enter a typical rate against each item. They were

also asked to note if the item was rated as a result of either:

detailed analysis (D); experience (E); "what the job will stand"

(S); or just ignore (I).

The estimators from the original interview sample of ten small

builders were asked to complete the questionnaire and nine

completed questionnaires were returned. One estimator

sub-divided his responses further into three stages to indicate

the different approaches he used. These depended upon whether

he was, a) keen, b) reasonably interested, or c) not

particularly interested, in obtaining the contract. The

requirements of the questionnaire had been explained previously

 13

in the informal interview with the estimators. All responses

were provided anonymously. It was felt that this would yield

beneficial results as most estimators were quite happy to

co-operate in the survey but were less than happy to divulge

extremely confidential information such as rates and labour

constants. It was also felt that the source of such rates and

labour constants would not be either relevant or appropriate

data as the intention was only to establish the method by which

estimators rated items in bills of quantities.

From a total of 324 item ratings, 173 were made by method D, 126

by method E, 14 by method S, and 11 by method I, ie., 53.4% of

items were built up in detail, 38.9% were analysed from

experience, 4.3% were "what the job would stand", 3.4% were

ignored. If these proportions are maintained for entire bills

of quantities, this would mean that just over one half the items

would be rated by the prescribed detailed method and just under

half the items would be rated without calculation. Whilst some

of the experientially derived rates are likely to be based on

sound judgement, these figures certainly indicate a significant

departure from the prescribed approach.

Analysis of questionnaire data

It is likely that there are many factors (including behavioural

factors) that determine an estimator's method of item rating.

Of these, the assertion recorded in the interviews - that

 14

detailed rate analysis was only undertaken for items of

significant value - was examined.

Fig 1 shows the relationship between the item total and the

number of estimators rating the item by detailed analysis.

Visual inspection of Fig 1 clearly indicates that the greater

the item total the greater the number of estimators in the

sample who used the detailed method of item rating.

To examine the extent to which these results may be generalised

to the population of smaller builders, a series of statistical

analyses were carried out to test the significance of the

observed trends. This was done (1) by bivariate analyses of the

correlation between the total number of estimators using method

D of item rating and the item total, (a) for all the items and

(b) by trade subsection and, (2) by a multivariate regression

analysis on the trade subsections simultaneously. As is common

in these kind of analyses, significance was judged at the

(conservative) 5% level for all the analyses described below.

Bivariate analysis. The estimated coefficient of correlation

between the frequency of the use of method D and the item total

was (i) for all items r=0.48 (n=37 p=0.001), (ii) for Groundwork

items r=0.43 (n=14 p=0.062), (iii) for In-situ Concrete items

r=0.34 (n=12 p=0.137) and (iv) for Masonry items r=0.64 (n=9

p=0.032) confirmed the significance of a general trend to more

detailed estimates with the higher valued items, except in-situ

concrete items and groundwork items.

 15

Multivariate analysis. A multiple linear regression analysis

was applied. The frequency of the use of method D was the

dependent variable and the item total was the independent

variable with the groundworks and in-situ concrete sections

being represented by dummy independent variables. This resulted

in a multiple r=0.73 (F=9.345 df=4,32 p=0.000) with the constant

and groundworks dummy coefficients significant at the 5% level.

This again confirms the significance of a trend to more

detailed estimates with the higher valued items except in-situ

concrete items.

Other models. Various other models were tested, including the

introduction of a squared term for the item totals and several

interaction terms. None of these was found to have a

significant improvement on the original model excluding the in-

situ concrete variable.

Introduction of further explanatory variables. Following

Runeson (1988) further potential explanatory variables were

examined. These comprised for each item: the proportion of

labour content; the quantity; the value contribution to the

total bill; the mean inter-estimator intra-item (RA) rate; the

RA standard deviation; the RA mean item total; the RA standard

deviation of the item total; the RA coefficient of variation;

the RA coefficient of skewness; and the RA coefficient of

kurtosis.
 16

Using Spon's price book (Spon, 1990) the proportion of labour

content was estimated for each item; the value contribution of

each item was taken as a proportion of the total value of all 36

items rated; and the RA coefficient of variation, skewness and

kurtosis is the same for both the item rates and item total.

The values for each of these further variables, together with

the number of detailed estimates recorded for the estimators,

for each of the 36 items are shown in Table 1. The top right

hand half of Table 2 summarises the significant correlations

found between the variables. As Table 2 shows, the number of

detailed estimates was correlated with the proportional labour

content of the items (r=-0.36), the value contribution of the

item (r=0.47), the RA mean item total (r=0.48), the RA total

(r=0.42) and the RA coefficient of variation (r=-0.66). However

the skewness and kurtosis were also correlated with each other

and the skewness was correlated with the RA coefficient of

variation.

In an attempt to find a simpler model, the logs of the item

rates and item totals were taken. The significant correlations

of these transformed variables are shown in the bottom left hand

half of Table 2. As can be seen the RA skewness and kurtosis

coefficients were now uncorrelated with the other variables,

leaving all the other variables, except item quantity,

correlated. As it was suspected that all these variables were

in some way related to the 'size' of the item, the partial

correlations were examined via the regression coefficients

 17

obtained by regressing the proportion of labour content, the

quantity, the value contribution to the total bill, the mean

inter-estimator intra-item (RA) log rate, the RA standard

deviation, the RA mean log item total, the RA standard deviation

of the item total, the RA coefficient of variation, the RA

coefficient of skewness and the RA coefficient of kurtosis on

the number of detailed estimates. This produced a regression

model with an adjusted r2 of 0.599 (p=0.000, SE=1.975) with only

one significant variable, that of the RA mean log item total

(beta=0.694). This confirmed the importance of item total in

determining the rating method used1.

From these analyses, it was therefore concluded that there was

sufficient evidence to accept the proposition that a simple

linear trend exists between item total and frequency of detailed

estimating for all the trade sections except in-situ concrete

work.

Discussion

These results confirm the fact that smaller builders' estimators

use methods other than the prescribed detailed rate analysis and

suggest that estimators apply such non-detailed methods for up

1
As a matter of interest, these variables were also
regressed on the transformed data standard deviations (the
results are the same for both item rate and total item values).
This produced an adjusted r2 of 0.440 (p=0.001, SE=0.173) with
only the mean item rate variable being significant (beta=-0.595)
 18

to 50 percent of bill items, certainly for the work sections

studied. Although the statistical relationships found in this

present study were not always strong enough to be absolutely

conclusive, there is clear evidence of a direct relationship

between the method of item rating and item total for each of the

three work sections examined. The expected positive correlation

between the frequency of use of the prescribed detailed rating

method across estimators and item total, was found for each work

section although the result was not always statistically

significant.

Of the nine questionnaires returned, all nine agreed that the

item total was a major factor when deciding on whether or not to

analyse a rate for an item in detail. Note that it is the total

value of the item, i.e., unit rate multiplied by quantity, and

NOT the value of the unit rate itself that is being considered

here. When the estimators were requested to state what they

considered a significant item total, one suggested £1,000, one

£2,000, the remainder declined to answer. All nine respondents

agreed that the significant amount was not a static amount but

varied according to the total estimated value of the contract

under consideration. Differences between estimators were found

in the critical item total value, the determination of which

seemed to be the result of a subjective judgement made by the

estimator. Thus it appears that it is this application of

judgement that causes the different methods of rating of the

same item between estimators. It is likely therefore that

behavioural and environmental factors such as personality,

 19

motivation, incentive and habit, all influence the use of

experiential based estimating techniques particularly when

dealing with items considered less than a critical value.

Clearly, the factors that influence this judgement are worthy of

further research.

The methods of rating bills of quantities items, identified in

the survey, are clearly designed to allow the estimator to

concentrate on those items that he considers to have significant

influence on the final value of the estimate. The prescribed

detailed method of item rating is time consuming and tender

periods are short. It would appear that the combination of item

rating methods is perceived to be the most efficient way for the

estimator to undertake his task and arrive at a reasonably

accurate tender price in the time available and with minimum

opportunity cost.

The interviews clearly revealed that the estimators perceived

the prescribed method of detailed item rating to be the most

accurate method of estimating. The implication, therefore, is

that using more subjective methods is likely to be less

accurate. If this is the case, it is expected that the

difference between the item ratings and the resulting actual

costs associated with each item will be greater when more

subjective methods are used. Whilst the data for such an

analysis are not generally available, either because of

confidentiality or insufficiently detailed accounting practices,

this variability can be estimated indirectly by recourse to the

 20

method normally employed in PERT. Here, the estimator is

required to provide a 'likely maximum' and a 'likely minimum'

cost, these being essentially subjective estimates of the range

of differences between estimated and actual costs. To explore

this further, an additional survey was undertaken with the

estimators, in an attempt to identify the range associated with

the intra-estimator intra-item rates and this is described in

the following section.

THE RANGE OF INTRA-ESTIMATOR INTRA-ITEM RATES

Introduction

Apparently item rates may vary considerably both within and

between estimators for a variety of reasons. Little research

has been undertaken to establish the range of intra-item rates

applied by each estimator. Most researchers agree that no

individual unit rate is 'correct', but that an estimator's rate

is one of a family of such rates, any one of which could be

correct (eg., Beeston, 1983; Fine, 1987). On this basis

therefore the intra-estimator intra-item (AA) range or

variability of rates was taken as a surrogate for degree of

correctness or accuracy of the rate.

Barnes (1971:A.3.5) attempted to analyse the effect of different

accuracies of item ratings on the accuracy of the total estimate

by a measurable 'accuracy ratio' and concluded that "attention

 21

paid to pricing smaller value items is wasted in that it has

negligible effect on the accuracy of the total." In developing

his accuracy ratio theory he states that "the estimation of unit

costs of the items is assumed to have a constant coefficient of

variance [sic]"(1971 p A2.2). The importance of this assumption

cannot be understated as the research undertaken by Barnes

eventually resulted in both the civil engineering and building

standard methods of measurement being revised to reduce the

number of items contained in bills of quantities, supposedly

without significantly affecting the accuracy of the total tender

figure.

The first question of interest concerns Barnes' assumption of

homogeneity of AA variability. This can be approached either by

applying one of the usual homogeneity tests, such as Bartlett's

test, or by testing for the existence of trends, such as a

correlation between item variability and item total or rating.

In the work described next, the AA variability is analysed in

terms of both standard deviations and coefficients of variation.

Data

The ten estimators were sent a further questionnaire comprising

the same items as in the previous study (Appendix A - see

italics). They were requested to insert against each item:-

a) lowest rating they would consider using (ie., proxy for

 22

minimum expected cost)

b) 'normal' rating (ie., proxy for mean expected cost)

c) highest rating they would consider using (ie., proxy for

maximum expected cost)

These questionnaires were administered in March 1991. Seven

estimators completed them fully and a further estimator

completed the groundwork section only. All the questionnaires

were completed anonymously. No further discussions were

possible with the estimators on specific information arising

from the questionnaire.

Analysis 1

It was clear from visual inspection of the unit rates that in

virtually every case the normal rate is closer to the minimum

rate than the maximum rate, ie., there is a positively skewed

distribution - a normal occurrence for these kind of data.

Therefore the standard deviation (see Appendix B) was estimated

by the PERT approximation for Beta distributions (Loomba, 1978):

b-a
σ=
6
 23

where σ = estimated standard deviation

b = maximum value

a = minimum value

The data included an item for steel reinforcing that was rated

by all estimators at a comparatively high rate of many hundreds

of pounds, whilst most items varied in value between a few pence

and 70 or 80 pounds. To remove the possibility of any undue

distortion, the analyses were conducted both with and without

this outlier.

The item ratings and their estimated AA standard deviations were

plotted for each estimator both with and without the outlier.

Visual inspection suggested that the AA standard deviation

generally increases as the item rate increases. The regression

line of item ratings on the AA standard deviation was also

plotted. The slopes, together with the correlation coefficients

both with and without the outlier, are summarised in Table 3.

As can be seen from Table 3, the presence of the outlier did

cause some apparent distortion of the results for estimators 2

and 6. However in the remaining cases there was a statistically

significant correlation between the item ratings and their AA

standard deviations. In both cases where the results were not

significant, the estimators had attributed a large number of

items with an AA standard deviation between 0 and 1. The

remaining items had a limited spread of AA standard deviation,

 24

yet estimator 2 had one item with an AA standard deviation of

171, the highest recorded AA standard deviation of any rate by

any estimator. It would appear that both estimators 2 and 6

limit their range to a very large extent, with estimator 6

having 21 out of 36 ratings with a zero AA standard deviation.

A reliability analysis was carried out to check on whether the

observed differences in AA standard deviations between

estimators were 'real' rather than just simply artifacts of the

sampling process. This was done for the seven estimators for

which a full set of data was available. This analysis produced

a Cronbach alpha value of 0.6618 (standardised Cronbach alpha =

0.9453) with an average inter-estimator correlation of 0.8381.

The lack of any probability levels associated with the Cronbach

alpha statistic makes interpretation of this result rather

subjective, but this was taken to indicate a reasonable level of

reliability. As a further check, a components of variance

analysis was made (Table 4). This indicated conclusively that

there were no significant differences between the estimators.

Thus it was decided to continue the analysis with the data

pooled across the estimators, ie., all the estimators were

treated as if they behaved in the same way.

Having pooled the data, the first test was to check the

differences between AA standard deviations. A oneway ANOVA

showed the differences to be statistically significant

(F35,231=3.1177, p=0.0000; F34,225=2.4535, p=0.0001 with outlier

removed). The next test examined the significance of the

 25

association between the AA standard deviations and their AA mean

and total. A two-way analysis of covariance (ANCOVA), where all

'effects' were adjusted concurrently, indicated the existence of

significant item effects (F=1.88, p=0.003; F=4.326, p=0.000 with

outlier removed) and AA means effects (F=241.4, p=0.000;

F=240.8, p=0.000 with outlier removed). It did not indicate the

existance of significant of item totals (F=0.027, p=0.602)

except with the outlier removed (F=0.001, p=0.974). A further

two-way ANCOVA indicated the existence of work section effects

(F=4.336, p=0.014) and AA means effects (F=221.818, p=0.000) but

not with item totals (F=26.843, p=0.642). Of course the nature

of the data collected was such that it was not possible to test

if the work section effect subsumed any item effects.

The results of regressing AA means and item totals on the AA

standard deviations are summarised in Table 5. This table

confirms the significance of the slopes of the regression lines

for each work section and the correlation of AA standard

deviations with AA means and not item totals.

Discussion

The AA standard deviation is a measure of the dispersion of

values around the AA mean, and can be taken to represent the

family of rates for each item. The correlation between AA

standard deviation and item ratings is not surprising as it is

usual for researchers in this field to assume the existence of a

 26

proportional relationship, and hence the general use of

coefficients of variation. What is a little surprising however,

is the lack of any significant correlation between AA standard

deviation and item total, ie., the AA standard deviation is

unaffected by the quantity of work represented by the item.

Having established the existence of a 'size' effect in using AA

standard deviation as a measure of variability, attention was

next turned to examining more closely the nature of this

relationship and, in particular, the possible neutralising

effects of using AA coefficients of variation as a substitute

measure of variability.

Analysis 2

The same procedure as Analysis 1 was followed using the AA

coefficient of variation in lieu of the AA standard deviation.

Table 6 summarises the results. Statistically significant

correlations were obtained in only two cases out of fifteen.

The Cronbach alpha value was 0.3870 (standardised Cronbach alpha

= 0.3201) with an average inter-estimator correlation of 0.0743,

suggesting quite poor consistency between the estimators. The

components of variance analysis (Table 7), indicated that there

were no significant differences between the estimators. Again

it was decided to continue the analysis with the data pooled

across the estimators ie. all the estimators were treated as if

 27

they behaved in the same way.

Having pooled the data, the first test was to check the

differences between AA coefficients of variation. A oneway

ANOVA indicated the differences to be statistically significant

(F35,231=1.478, p=0.0489; F34,225=1.503, p=0.0443 with outlier

removed). The next test was to find if the differences between

AA coefficients of variation were significantly associated with

differences between their AA mean and total values. A two-way

analysis of covariance (ANCOVA) indicated the existence of

significant item 'effects' (F=1.57, p=0.028; F=1.634, p=0.020

with outlier removed) and AA means effects (F=1.570, p=0.028;

F=1.634, p=0.020 with outlier removed) but not with item totals

(F=0.005, p=0.941; F=0.008, p=0.929 with outlier removed). A

further two-way ANCOVA indicated the existence of work section

effects (F=7.138, p=0.001; F=4.796, p=0.009 with outlier

removed) but no AA means effects (F=1.146, p=0.285; F=0.008,

p=0.927 with outlier removed) or with item totals (F=0.431,

p=0.512; F=0.377, p=0.540 with outlier removed). Again, the

nature of the data collected was such that it was not possible

to test if the work section effect subsumed any item effects.

What is clear from this analysis is that there are no linear

item size effects once the work section effects are removed.

The possibility remains however that the relationship between

the item size and AA coefficients of variation differs between

sections (the ANCOVA analysis not including interaction terms

due to the limited amount of data available).

 28

The results of regressing AA means and item totals on the AA

coefficients of variation by work sections are summarised in

Table 8. This table indicates the significance of the slopes of

the regression lines for each individual work section and the

significant correlation of AA coefficients of variation with AA

means and not item totals.

Overall, these results indicate that there is no significant

linear relationship between AA coefficients of variation and

item totals for each or all of the work sections. There is a

significant positive linear relationship between the AA

coefficients of variation and the AA means for the ground works

and concrete works sections.

Barnes' Assumption

In view of its importance, the assumption by Barnes that all

items have a constant AA coefficient of variation was examined

in more detail.

The AA coefficients of variation for all rates submitted by the

estimators are categorised in Table 9. This suggests that,

although the assumption proposed by Barnes is not statistically

valid according to the above, it may not be an unreasonable

proposition particularly when the following is considered. Out

of 267 items, a total of 21 have an AA coefficient of variation

 29

greater than 10. Of these, 6 of the items related to earthwork

support or disposal of water, which can reasonably be considered

high risk items with extremes of rates available - a high AA

coefficient of variation in such items is therefore not

unreasonable. Most of the remaining items with an AA

coefficient of variation in excess of 10% had very high maximum

rates applied to them which tended to distort the AA coefficient

of variation.

Those items in the band over 5% and not exceeding 10% AA

coefficient of variation totalled 50. Of these, 31 exceeded 5%

but did not exceed 6% AA coefficient of variation with a further

6 in the 6% to 7% AA coefficient of variation range. Of these

items it is the maximum rate that distorts the figures and

increases the AA coefficient of variation. Considering how

infrequently an estimator gets the opportunity to apply the

maximum rate and how much more frequently rates applied will be

somewhere between the minimum and normal, Barnes' assumption

appears to be more reasonable.

Barnes' hypothesis, that "attention paid to pricing smaller

value items is wasted in that it has negligible effect on the

accuracy of the total", was now tested. Barnes based his

findings on data from rated bills of quantities, these showing

final tender figures and a figure for each individual rate.

Should several bills of quantities be available for the same

project a comparison could be made between the figures inserted

in the bills by each estimator for each item. However each

 30

estimator would insert only one rate against each item, thus

Barnes could not have known the AA variability.

The data produced by this research are an indication of the AA

variability for each of 36 bill items.

The coefficient of variation of the total of the 36 items is

∑ qi2 S i2
Vt =
∑ q i xi

Vt = Variability of total

qi = quantity of item

Si = AA standard deviation of item

xi = AA mean of item

As qi is known and both Si and xi are estimated from the data it

is possible to calculate Vt. If the hypothesis proposed by

Barnes is correct then the greatest effect on the variability of

the total figure is caused by the high value items/rates with

lower value items/rates having progressively less effect as they

reduce in value. The 36 items were therefore rank ordered by

item total value and Vt calculated for sets of 1, 2, ... , 36

items. This was repeated for all the estimators and the results

are shown in Fig 2. This shows that, in virtually every case,

Vt reduces very rapidly to a near minimum with only a few of the

highest rated items. This appears to be the case no matter what

 31

starting level of variability is considered. Thus, if the

intra-item variability measure is equated with Barnes' intra-

item accuracy, these results would appear to confirm the notion

that many small value items have a negligible effect on the

accuracy of the final figure. This fact lends some support to

the hypothesis that detailed attention paid to rating many small

value items is likely to be wasted because of the excessive

opportunity costs involved.

Conclusion

1. The relationship between the value of the unit rate and the

extent of the family of rates as measured by the AA standard

deviation was found in the majority (13 out of 15) of cases

to be significant. As one would expect, high value items

tend to have a higher AA standard deviation than low value

items. This indicates that the range of rates available for

such items is more extensive than for low value items. This

would support the viewpoint that estimators make the most

beneficial use of their time without reducing estimate

accuracy by concentrating their efforts on high value items.

2. The assumption by Barnes, that all items have a constant AA

coefficient of variation is reasonable. The vast majority of

items has an AA coefficient of variation of less than 7%,

with 73% of items examined having an AA coefficient of

variation of less than 5%.

 32

3. The hypothesis by Barnes that little attention is paid to

rating small value items is supported by the results for each

of the eight builders' estimators examined. In each case the

total cost variability reduced rapidly with the items of

significant cost and did not appear to reduce further when

the items of minor cost were considered.

SUMMARY AND CONCLUSIONS

This paper describes a study of the estimating processes of

smaller builders. The study comprised firstly, a series of

interviews revealing four major differences between the standard

texts and practice: (1) limiting of estimating effort and cost

in tender preparation by using experience to rate certain items

rather than detailed rate analysis; (2) rating items inclusive

of profit and overheads; (3) limiting the cost of tender

preparation by reducing pre-tender planning and risk appraisal

to an absolute minimum; and (4) rating of preliminary items on a

basis of what might eventually be wanted rather than what was

specified in the tender documents. This was followed in the

second part of the study by a questionnaire survey concerning

the estimators' approach to rating bills of quantities items.

An experiment involving eight builders' estimators separately

rating an extract from a bill of quantities containing 36 items,

is described from which it was found that only just over half of

the items were rated by the detailed methods prescribed in the

 33

standard texts. The remaining items were rated mainly by

'experience'. In attempting to identify which items are more

likely to be rated by detailed methods, several potential

variables were analysed including the work section containing

the item, the item rate, the item quantity, the item total, the

labour content of the item, the contribution of the item total

to the total of the bill, the RA standard deviation of rates,

the RA standard deviation of item totals, the RA coefficients of

variation, skewness and kurtosis. The result shows that the

item total (rate multiplied by quantity) is the one significant

determining variable, although the degree of influence of this

differs between work sections.

In the final part of the study the intra estimator variability

of item rating was examined by questionnaire survey through the

estimators' admitted range of rating values. On the assumption

that this reflected the range of accuracy with which the items

are rated (costed), Barnes' assumption of constant AA

coefficient of variation was tested and found to be reasonable.

It is considered that this work is important in contributing to

the 'bottom-up' understanding of the mechanisms underlying

movements of building prices at both project, market and

industry levels. At the project level, the economic

consequences of design decisions are embodied in builders'

pricing behaviour. As the nature of most procurement practises

is to delay the builders' involvement until after the major

design decisions are made, it is necessary to understand, and

 34

therefore predict, such behaviour in advance. Similarly, the

aggregated builders' pricing behaviour has an impact at both

market and industry levels. By modelling this behaviour, it may

be possible to forecast in advance the economic consequences of

design decisions for individual projects and ultimately changes

in design policies for markets and the industry (eg changes in

building regulations and/or design codes). A further and more

immediate application is in the formulation of measurement

codes. By considering the pricing practises of smaller

builders, it should be possible to gauge the benefit/cost

effects of new approaches.

ACKNOWLEDGEMENTS

Credit should be given to David and Karen Eaton, the anonymous

reviewers and the editors for their kind and helpful comments in

the preparation of the final version of this paper.

REFERENCES

Barnes, N.M.L., 1971, The design and use of experimental bills

of quantities for civil engineering contracts, PhD thesis,

University of Manchester Institute of Science and Technology.

Beeston, D.T., 1983, Statistical methods for building price

data, E & F N Spon, ISBN 0419 12270 2.

 35

Eastham, R.A., 1990, The decision to tender within current

contractual arrangements, Proceedings CIB W-92 Symposium on

Procurement Systems, Zagreb, Yugoslavia, 10-13 September

Institute Council for Building Research, 67-77.

Fine, B., 1974, Tendering strategy, Building, 25 Oct 115-21.

Fine, B., 1987, Kitchen sink economics and the construction

industry, in Building Cost Modelling and Computers ed P.S.

Brandon, E & F N Spon, ISBN 0 419 14040 9, 25-30.

Hampson, J., 1979, Estimating procedure utilised in medium

sized contracting organisation, undergraduate dissertation,

Department of Civil Engineering, University of Salford.

Housing and Construction Statistics 1977-1987, H.M.S.O.,

Department of the Environment.

Key Statistics for Urban Areas, 1981, The north cities and

towns, H.M.S.O., Department of the Environment.

Loomba, N.P., 1978, Management - a quantitative perspective,

Collier Macmillan International, ISBN 0 02 371640 1.

Runeson, G., 1988, An analysis of the accuracy of estimating and

the distribution of tenders, Construction Management and

Economics, vol 6, 357-70.

 36

Spon, 1990, Building Price Book, E & F N Spon.

 37

Appendix A: Sample items

Local Authority School; floor area 500m2; location within 5 miles of head
office.
Bill of Quantities prepared in accordance with SMM7.

Please look at the following extracts from the Bill of Quantities for the
above. If you were tendering for the contract, which items would you price
in detail (please mark these D), which would you price from experience
(please mark E), which items would you ignore (please mark I), and which you
would price for what the job will stand (S).

Please insert in the appropriate column:

a) lowest unit rate you would generally consider using
b) normal rate ie., one generally used

c) highest rate you would generally consider using

Total responses

D E I S
Item.
GROUNDWORK
Excavating and filling

1. Excavating topsoil for preservation average depth

150mm 400 m2 3 5 1 0

2. Excavating to reduce levels; maximum depth not

exceeding 0.25m 10 m3 3 5 1 0

3. Ditto not exceeding 1.00m 300 m3 5 3 1 0

4. Excavating basements and the like; maximum depth

not exceeding 2.00m 50 m3 5 3 1 0
3
5. Excavating pits (12 Nr) ditto 10 m 3 5 1 0

6. Excavating trenches; width over 300; maximum

depth not exceeding 1.00m 225 m3 3 5 1 0

7. Working space allowance to excavations; reduce

levels; basements or the like 30 m2 5 2 2 0
8. Earthwork support; maximum depth not exceeding
1.00m; distance between opposing faces not
exceeding 2.00m 200 m2 0 4 4 1
9. Disposal; surface water Item 0 2 1 6
3
10. Disposal excavated material off site 555 m 6 3 0 0
3
11. Ditto on site 100 m 5 4 0 0
12. Filling to excavations, average thickness not
exceeding 0.25m arising from excavations 60 m3 3 6 0 0

13. Ditto obtained off site; clean broken stone 60 m3 5 4 0 0

14. Surface treatment; compacting; filling;
 38

blinding with sand 400 m2 3 4 0 2

15. Surface treatment; compacting; bottoms of
excavations 400 m2 2 5 1 1

IN-SITU CONCRETE/LARGE PRECAST CONCRETE

Plain in-situ concrete; BS 5328; designed mix
C25; 20 aggregate

16. Foundations 100 m3 7 2 0 0

17. Ground beams 5 m3 7 2 0 0
18. Filling to hollow walls 5 m3 6 3 0 0
Reinforced in-situ concrete; BS5328;
designed mix C35; 20 aggregate; vibrated

19. Beds; thickness not exceeding 150mm 10 m3 8 1 0 0

20. Beds; thickness 150-450mm 80 m3 8 1 0 0

21. Beds; thickness over 450mm 20 m3 8 1 0 0

22. Columns 10 m3 7 2 0 0

23. Staircases 5 m3 7 2 0 0
Formwork for in-situ concrete

24. Sides of foundations; plain vertical height

250-500mm 200 m 7 2 0 0

25. Soffits of slabs; slab thickness not exceeding

200mm; horizontal 15 m2 6 3 0 0
Reinforcement for in-situ concrete

26. Bar; 12mm diameter; straight 0.5 t 5 4 0 0

Worked finishes

27. Power floating 400 m2 3 6 0 0

MASONRY
Concrete commons in cement mortar (1:3)

28. Walls; half brick thick; vertical 20 m2 7 2 0 0

Nori red rustic facings in cement mortar
(1:3)

29. Walls; half brick thick; fair face and

struck jointed one side; vertical 600 m2 9 0 0 0

30. Walls; one brick thick; one face in

facings fair face and struck jointed;
one face in concrete commons fair face
and struck jointed; vertical 50 m2 9 0 0 0
Celcon concrete blockwork in cement
mortar (1:3)
 39

31. Walls; 100 thick; vertical 590 m2 8 1 0 0

32. Closing cavities 50 wide with 100
blockwork; vertical 50 m 3 6 0 0
Accessories/sundry items

33. Forming cavities; in hollow walls;

50 wide; butterfly wall ties 5Nr/m2 595 m2 1 7 0 1
34. Damp proof courses; hyload; width not
exceeding 225mm; vertical 25 m 2 7 0 0

35. Ditto horizontal 195 m 2 7 0 0

36. Ditto raking 10 m 2 7 0 0
-------------
Total 173 126 14 11
=============
 40

Appendix B: Item rates

───────────────────────────────────────────────────────────────────────────
Item Estimator
1 2 3 4 5 6 7 8
───────────────────────────────────────────────────────────────────────────
1 0.18 1.18 0.28 0.20 0.62 0.19 0.30 0.53
0.012 0.408 0.005 0.007 0.013 0.023 0.008 0.028
2 1.20 2.20 1.58 1.66 3.39 0.74 2.00 6.50
0.083 0.167 0.038 0.168 0.077 0.012 0.067 0.353
3 0.95 2.20 1.58 1.46 2.54 0.74 2.00 2.75
0.083 0.167 0.038 0.167 0.058 0.012 0.067 0.150
4 3.50 6.50 5.70 6.10 2.54 8.99 6.67 7.50
0.167 0.333 0.142 0.342 0.058 0.143 0.183 0.407
5 5.25 11.00 7.12 7.70 10.15 4.10 11.05 7.50
0.208 0.375 0.170 0.348 0.232 0.065 2.197 0.402
6 4.95 12.00 4.75 5.14 5.07 4.95 7.78 4.50
0.217 0.333 0.113 0.100 0.117 0.078 0.183 0.250
7 3.75 12.00 22.75 2.46 6.40 9.44 13.82 12.50
0.150 0.333 0.400 0.123 0.137 0.150 0.175 0.678
8 0.50 3.50 0.60 1.10 1.26 5.89 3.03 3.70
0.117 0.600 0.015 0.058 0.038 0.093 3.047 0.202
9 0.00 0.00 100.00 0.00 0.00 100.00 0.00 0.00
0.000 16.667 41.667 0.000 0.000 12.500 0.000 0.000
10 6.50 7.00 6.00 5.58 7.70 9.14 6.15 8.20
0.442 0.333 0.333 0.320 0.117 0.145 0.900 0.443
11 2.20 2.10 3.00 1.16 2.20 1.42 2.94 4.20
0.167 0.508 0.333 0.042 0.033 0.023 0.032 0.227
12 3.00 3.89 1.42 2.76 3.39 6.66 5.80 15.00
0.167 0.185 0.083 0.057 0.077 0.105 0.000 0.812
13 13.50 12.00 14.50 16.16 16.57 19.53 14.48 15.00
0.692 0.242 0.367 0.577 0.367 0.310 1.368 0.812
14 0.45 0.80 0.60 1.92 0.76 1.29 1.12 1.20
0.033 0.077 0.043 0.077 0.023 0.020 0.000 0.065
15 0.25 0.50 0.05 0.34 0.27 0.37 0.67 0.60
0.033 0.117 0.000 0.010 0.010 0.007 0.103 0.032
16 51.00 60.00 58.58 59.96 57.88 68.62 53.94
1.000 3.200 1.903 1.475 1.475 1.423 0.000
17 52.50 65.00 65.00 62.69 64.96 126.00 59.36
0.833 2.500 1.667 1.363 1.738 2.000 0.000
18 55.00 55.00 75.00 63.38 116.88 76.65 66.32
1.083 2.833 1.167 1.088 3.625 1.217 0.000
19 57.50 68.00 65.00 69.72 68.51 75.06 65.76
0.917 0.500 1.167 1.283 1.868 1.192 0.000
20 55.00 64.00 64.00 67.60 61.43 75.06 60.24
1.000 0.667 1.667 1.432 1.607 1.192 0.000
21 52.50 58.00 62.00 67.60 54.34 73.15 56.56
0.917 0.000 1.333 1.432 1.343 1.147 0.000
22 57.50 125.00 75.00 72.60 86.23 77.66 109.02
0.833 0.000 1.250 5.492 2.525 1.233 0.000
23 70.00 132.00 75.00 72.60 100.40 67.34 74.96
1.333 0.000 1.250 5.492 3.050 1.068 0.000
24 5.50 5.50 5.00 5.38 14.53 20.21 14.76
0.417 0.292 0.417 0.663 0.495 0.322 0.000
25 18.75 7.00 45.00 22.38 7.55 29.65 17.36
0.750 0.210 1.167 1.450 0.255 0.470 0.000
26 495.00 1000.00 360.00 601.00 663.80 722.09 872.50
12.500 171.667 16.667 37.333 17.900 11.462 7.352
27 1.60 1.00 3.00 1.32 1.35 2.78 1.50
0.058 0.000 0.167 0.032 0.050 0.045 0.187
28 10.50 25.00 18.20 20.38 17.26 19.29 20.05
0.250 1.333 0.450 0.977 0.530 0.307 0.000
29 30.75 35.00 53.07 36.72 44.48 31.53 39.52
0.500 1.067 1.150 1.090 1.190 0.500 0.000
30 41.25 60.00 84.53 62.74 61.33 51.77 60.03
0.750 3.400 2.167 1.313 1.705 0.822 0.000
31 10.75 24.00 18.00 17.79 20.95 10.44 19.80
 41

0.250 0.667 1.000 0.387 0.600 0.165 0.098

32 1.00 4.00 1.20 4.96 1.08 2.04 4.39
0.117 0.400 0.067 0.147 0.040 0.032 0.000
33 0.75 3.00 0.50 1.39 1.00 1.15 0.69
0.067 0.167 0.008 0.075 0.030 0.018 0.000
34 1.50 2.50 0.76 1.49 1.92 2.76 1.66
0.050 0.333 0.013 0.028 0.050 0.043 0.000
35 1.50 1.50 0.72 1.49 2.19 1.42 1.53
0.050 0.167 0.010 0.028 0.060 0.023 0.000
36 1.60 3.50 0.76 1.49 1.26 2.10 1.53
0.050 0.342 0.017 0.028 0.038 0.033 0.000
───────────────────────────────────────────────────────────────────────────S
tandard deviations given in italics.
 42

─────────────────────────────────────────────────────────────────────────────────────
───────────────────
Item Lab Q cntrib mean SD mean SD COV skew1
kurtosis
(%) (%) rate total
─────────────────────────────────────────────────────────────────────────────────────
───────────────────
1 22 400 0.46 0.43 0.3424 174.00 136.9713 78.72 1.76 3.11
2 22 10 0.10 2.41 1.8285 24.09 18.2855 75.91 1.93 4.05
3 22 300 1.57 1.78 0.7227 533.25 216.8125 40.66 -0.14 -1.25
4 20 50 0.80 5.94 2.0764 296.88 103.8188 34.97 -0.44 -0.07
5 20 10 0.19 7.98 2.5859 79.84 25.8595 32.39 -0.15 -1.24
6 20 225 3.06 6.14 2.5809 1382.06 580.7065 42.02 2.13 4.39
7 20 30 0.81 10.39 6.4959 311.70 194.8766 62.52 0.75 0.81
8 72 200 1.43 2.45 1.9011 489.50 380.2146 77.67 0.73 -0.29
9 - - 0.04 25.00 46.2910 25.00 46.2910 185.16 1.44 0.00
10 - 555 10.27 7.03 1.2241 3903.73 679.3898 17.40 0.64 -0.62
11 53 100 0.76 2.40 0.9680 240.25 96.8028 40.29 0.69 0.56
12 47 60 1.32 5.24 4.2861 314.40 257.1644 81.80 2.04 4.62
13 18 60 2.25 15.22 2.2591 913.05 135.5434 14.85 0.73 1.24
14 28 400 1.07 1.02 0.4694 407.00 187.7719 46.14 0.88 0.83
15 100 400 0.46 0.38 0.2023 152.50 80.9215 53.06 -0.09 -0.41
16 17 100 9.06 58.57 5.5407 5856.86 554.0720 9.46 0.67 1.51
17 21 5 0.55 70.79 24.7654 353.94 123.8272 34.99 2.45 6.30
18 36 5 0.55 72.60 21.3184 363.02 106.5920 29.36 1.80 3.74
19 37 10 1.04 67.08 5.3446 670.79 53.4457 7.97 -0.55 1.78
20 32 80 7.91 63.90 6.2844 5112.34 502.7519 9.83 0.63 1.24
21 29 20 1.88 60.59 7.4871 1211.86 149.7423 12.36 0.82 -0.44
22 43 10 1.31 86.14 23.2135 861.44 232.1348 26.95 0.80 -0.13
23 49 5 0.65 84.61 23.5816 423.07 117.9081 27.87 1.75 2.56
24 74 200 3.10 10.13 6.2471 2025.14 1249.4169 61.70 0.74 -1.33
25 68 15 0.49 21.10 13.2154 316.48 198.2303 62.64 0.90 0.80
26 27 0 0.52 673.48 217.5849 336.74 108.7924 32.31 0.13 -0.48
27 71 400 1.11 1.79 0.7749 717.14 309.9653 43.22 1.00 -0.79
28 59 20 0.57 18.67 4.3635 373.37 87.2696 23.37 -0.80 2.42
29 41 600 35.49 38.72 7.8824 23234.57 4729.4526 20.36 1.05 0.69
30 38 50 4.58 60.24 13.1016 3011.79 655.0778 21.75 0.72 2.26
31 50 590 15.54 17.39 5.0843 10260.10 2999.7115 29.24 -0.50 -0.94
32 86 50 0.20 2.67 1.7246 133.36 86.2316 64.66 0.34 -2.32
33 82 595 1.10 1.21 0.8436 720.80 501.9608 69.64 1.99 4.38
34 67 25 0.07 1.80 0.6725 44.96 16.8117 37.39 0.04 -0.20
35 58 195 0.45 1.48 0.4260 288.32 83.0635 28.81 -0.23 2.92
36 75 10 0.03 1.75 0.8702 17.49 8.7022 49.77 1.52 3.10
─────────────────────────────────────────────────────────────────────────────────────
───────────────────
1
Significant values in bold
Table 1: Item variables
 43

Lab Q Contrib Mean SD Mean SD COV Skew Kurt Detailed rate

rate total

Lab 1.00 0.36 -0.36

Q 1.00 0.55 0.51 0.59

Contrib 0.54 1.00 0.99 0.95 0.47

Mean rate -0.41 1.00 0.97

SD 0.47 0.46 -0.68 1.00

Mean total 0.72 0.42 -0.46 1.00 0.95 0.48

SD 0.47 0.46 -0.68 1.00 -0.46 1.00 0.42

COV 0.34 1.00 0.36 -0.66

Skew 1.00 0.65

Kurt 1.00

Detailed rate -0.36 0.47 0.64 -0.52 0.68 -0.52 1.00

Table 2: Correlations
 44

Estimators

1 2 3 4 5 6 7 8

Including outlier

Slope 0.02 0.14 0.06 0.06 0.03 0.01 0.02 0.05

Calculated r 0.993 0.975 0.651 0.980 1.000 0.843 0.748 1.000

Crit r @ 95% conf 0.325 0.325 0.325 0.325 0.325 0.325 0.325 0.325

Crit r @ 99% conf 0.418 0.418 0.418 0.418 0.418 0.418 0.418 0.623

Excluding outlier

Slope 0.02 0.01 0.10 0.04 0.03 0.00 0.03 -

Calculated r 0.923 0.061 0.541 0.768 0.995 0.184 0.583 -

Crit r @ 95% conf 0.330 0.330 0.330 0.300 0.330 0.330 0.330 -

Crit r @ 99% conf 0.424 0.424 0.424 0.424 0.424 0.424 0.424 -

Table 3: Summary of calculated and critical correlation

coefficients for the two variables of unit rate and AA
 45

Sums of Degrees Mean F ratio Probability

squares of square
freedom
Between items 10753.4 35 307.2
Within items 22557.1 216 104.4
Between 735.4 6 122.6 1.179 0.318
estimators
Residual 21822.4 210 103.9
Total 33311.2 251

Table 4: Components of variance analysis

 46

Model multiple r (prob) constant (prob) slope (prob)

All
sections
Mean 0.675 0.000 -0.920 0.088 0.067 0.000
Outlier 0.365 0.000 -0.040 0.884 0.034 0.000
removed
Total 0.026 ns
Outlier 0.011 ns
removed
Work
section 1
Mean 0.829 0.000 -0.892 0.000 0.249 0.000
Total 0.045 ns
Work
section 2
Mean 0.684 0.000 -3.267 0.069 0.071 0.000
Outlier 0.399 0.000 0.320 0.173 0.014 0.000
removed
Total 0.094 0.392
Outlier 0.043 0.710
removed
Work
section 3
Mean 0.780 0.000 0.026 0.682 0.023 0.000
-6
Total 0.330 0.008 0.287 0.001 0.265(10 ) 0.008

Table 5: Regression of AA means/item totals on AA standard

deviations by work section
 47

Estimators

1 2 3 4 5 6 7 8

All items

Calculated r 0.224 0.016 0.190 0.083 0.060 0.101 0.007 0.339

Crit r @ 95% conf 0.325 0.325 0.325 0.325 0.325 0.325 0.325 0.497

Crit r @ 99% conf 0.418 0.418 0.418 0.418 0.418 0.418 0.418 0.418

All items less max rate

Calculated r 0.479 0.219 0.353 0.150 0.160 0.245 0.161 -

Crit r @ 95% conf 0.330 0.330 0.330 0.330 0.330 0.330 0.330 -

Crit r @ 99% conf 0.424 0.424 0.424 0.424 0.424 0.424 0.424 -

Table 6: Summary of calculated and critical correlation

coefficients for the two variables of unit rate and
coefficients of variation

Sums of Degrees Mean F ratio Probability

squares of square
freedom
Between 0.322 35 0.009
items
Within 1.250 216 0.006
items
Between 0.064 6 0.011 1.902 0.082
estimators
Residual 1.186 210 0.006
Total 1.573 251

Table 7: Components of variance analysis

 48

Model multiple r (prob) constant (prob) slope (prob)

All
sections
Mean 0.004 0.948
Outlier 0.122 0.050
removed
Total 0.087 0.158
Outlier 0.086 0.164
removed
Work
section 1
Mean 0.201 0.028 0.053 0.000 0.002 0.028
Total 0.031 0.742
Work
section 2
Mean 0.246 0.024 0.024 0.000 3.88(10-5) 0.024
-5
Outlier 0.415 0.000 0.044 0.000 3.28(10 ) 0.000
removed
Total 0.111 0.316
Outlier 0.086 0.458
removed
Work
section 3
Mean 0.172 0.177
Total 0.163 0.202

Table 8: Regression of AA means/totals on coefficients of

variation by work section
 49

Estimators

1 2 3 4 5 6 7 8 Total % of
items
Coefficient
of variation
>0 and <2 11 7 10 7 3 25 34 1 98 36.7
>2 and <5 12 11 17 18 33 5 96 36.0
<5 and <10 10 10 8 9 1 14 52 19.5
>10 3 8 1 2 5 2 21 7.8
Total 36 36 36 36 36 36 36 15 267 100.0

Table 9: Range of AA coefficient of variation for estimators 1

to 8
50
 51

Fig 1: Regression line of builders undertaking detailed analysis and item total
52
 53

Fig 2: Variability of the total sum related to the number of items for contractors 1 to 8