ROAD RESEARCH LABORATORY Department of the Environment RRL REPORT LR 427 A REVIEW OF RURAL TRAFFIC-COUNTING METHODS IN DEVELOPING COUNTRIES by. J. D.G. F. Howe, MSC B.Tech. Tropical ~ction Road Research Laboratory Crowthorne, Berkshire. 1972 CONTENTS Abstract Page 1 1. 2. 3. 4. 5. 6. 7. 8. Introduction Methods for counting rural traffic in developing countries 2.1 Duration of counting 2.2 Frequency of counting 2.3 Timing of counts 2.4 The quantity to be estimated Kenya study of rural traffic flow Errors in estimates of ADT from sample counts 4.1 Sample testing 4.2 Desirable accuracy of estimates of traffic flow 4.3 Practical limits to sample duration Results 5.1 Traffic variability and ADT 5.2 Errors from random continuous counts of different durations 5.3 Repeated random samples 5.4 Sampling errors for individual months Discussion 6.1 Rural traffic counting in the United Kingdom 6.2 Rural traffic counting in the USA 6.3 Use of automatic traffic counters in developing countries Conclusions References oC CROWN COPYRIGHT 1972 1 2 2 2 2 3 3 4 4 5 5 6 6 7 7 8 8 8 9 9 10 10 Extracts from the text may be reproduced provided the source is ackno wledged A REVIEW OF RURAL TRAFFIC COUNTING METHODS IN DEVELOPING COUNTRIES ABSTRACT This Report reviews methods of rural traffic counting currently used in developing countries and examines the accuracy of the resulting flow estimates. The results of a questionnaire survey among a sample of developing countries suggest that decisions on the duration, frequency, and timing of counts are at present arbitrary. Consequently, estimated daily traffic flows can rarely be expected to lie within f 30 per cent of the true value averaged over the whole year. Athough repeating counts at intervals throughout the year increases the accuracy of traffic estimates, this is achieved only at a disproportionate increase in cost. It is concluded that for any appreciable increase in the accuracy of rural traffic estimates much more needs to be known about the magnitude and causes of the variations in flow. This requires that automatic traffic counters be used on a wider scale than at present. 1. INTRODUCTION This Report reviews the methods of rural traffic counting currently used in developing countries. The object of the review was to examine the accuracy of estimates of rural traffic flow resulting from the counting methods, and, if necessary, to make suggestions for their improvement. The Report is the first in a series that is considering the design of traffic survey methods suitable for developing countries, Information on traffic flow is needed for many purposes, h determining the appropriate standards of layout and design for particular roads, in allotting the resources for maintenance and improvement between the different roads in a network and in making general planning decisions on the development of transport systems. Existing methods of traffic counting provide information that is often inadequate and of doubtful accuracy and this research has been undertaken to establish the most economical methods of survey to produce adequate and reliable data. Information on the methods in use for counting rural traffic was obtained from a questionnaire circulated to a sample of developing countries in 1970. This was supplemented by information obtained during visits and from technical publications. The efficiency of the traffic counting methods was tested using continuous traffic measurements conducted by the Road Research bboratory at 38 sites in Kenya. Comparative data were also obtained from two sites in Nigeria and 30 of the sites used in the United Kingdom 50-point census for 19691. The Report first considers methods of traffic counting currently in use in developing countries. Next a brief description is given of the Kenya experiment followed by an outline of the method of crdculating errors in traffic counts. Sample testing is then described and the results presented and discussed. 2. METHODS FOR COUNTING RURAL TRAFFIC IN DEVELOPING COUNTRIES Appendix 1 summarises the information on methods for counting rural traffic in fourteen developing countries located in South America, Africa, the Middle East and the Far East. The information was correct at the time of the survey, but, as some countries are making changes in their traffic census methods, it may no longer be so in certain cases. Notwithstanding this, conclusions drawn from the results are probably still broady correct. The aspects of particular interest to this study are the duration, frequency, and timing of counts. 2.1 Duration of counting The duration of counting is standardised in about half the countries sampled. It ranges from a sin#e 8-hour count (Turkey) to a continuous count for 15 days (Ethiopia), although the latter is exceptional, being for special counts only. The most common period is 12 hours (6am - 6pm or 7am - 7pm) repeated for 5 or 7 days. The complex 8-hour count each day for three weeks, that is proposed for some stations in Thailand, can be regarded as an effective 24-hour count for 7 days. 2.2 Frequency of counting For ‘nationaS censuses (counts made annually over the entire country) the frequency of counting varies considerably from country to country. One country states that the frequency of counting is irregular, two countries state once a year, four countries state twice a year, three countries state three times a year, and one country states four times a year. Were the frequency is more than once a year, it is usually related to the number of major climatic seasons. 2.3 ~ming of counts The timing of counts is not generally standardised, although for ‘nationa~ censuses some countries specify broad wet and dry (or harvest) seasons when counts will, or will not, be made. Thus, in respect of their timing, traffic counts in developing countries can be considered as random samples. The period for counting, however many hours and days it comprises, is effectively a random selection in that any period in the year other than a few containing obviously unusual activities, such as Easter or Christmas, can be chosen. Even when, as in some of the national censuses, certain months or periods are specified, sampling is stall essentially random since there is no evidence that the period chosen is selected on the basis of a known pattern of seasonal variation. Mso, experience has shown that in practice these periods are, regrettably, rarely adhered to. In some developing countries, the purpose of traffic counts is not always clear. It might be to provide estimates of average dady traffic in the specified week, month, or year, or merely the average flow during the observation hours. 2.4 The quanti~ to be estimated Athough it is rarely explicitly stated, rural traffic counts usually attempt to measure average rather than peak usage. The commonest measure of average usage is the amount of daily traffic. However, the word ‘dafly’ sometimes refers to a period of less than 24 hours. In the United Kingdom2 rural traffic counts are taken to obtain the 16-hour (6am - 10pm), seven-day, average flow occurring in August. Au current counting systems based on ‘m-hour days’ (where m <24) suffer from a number of drawbacks. Since m varies so much between countries this suggests that the particular value chosen is arbitrary. Certainly it is not normally possible or meaningful to assign limits of error to the traffic estimates that result. Further, the ‘m-hour’ days are not natural periods of human activity such as the day, week or month. Thus, variations in traffic-flow characteristics, which can only add to estimation errors, are to be expected, e.g. the distribution of traffic through the hours of the day wfll vary with route characteristics: the distribution on a major trunk route carrying a large proportion of goods vehicles is urdikely to be the same as that on a farm-to-market road. In the USA, the term ‘daily’ traffic has its normal meaning: the flow of vehicles passing a given location in 24 consecutive hours. The basis of American traffic observations is the quantity ‘Average Annual hily Traffic’ (abbreviated to ADT), which is defined as the ‘Annual average number of vehicles during 24 consecutive hours that pass a particular point on the road over the period 365 days’3. This term would seem to have a number of advantages not shared by the various ‘m-hour’ days. It is unambiguous, readily understandable, and corresponds with a natural period of human activity. Thus it eliminates those problems associated with variations in the hourly distribution of traffic in different locations. However, the most important advantage of the ADT concept is that it enables statistical methods to be applied to the problem of rural traffic counting. Generally, it would seem to be the most logical basis for traffic observations and is the one used in this analysis. 3. KENYA STUDY OF RURAL TRAFFIC FLOW The objectives of the Kenya study were to provide the data necessary for a quantitative evahration of current traffic count methods, and to allow various new counting strategies to be tested. The study can be separated into two stages: (i) the measurement of the total pattern of traffic variation for a full year and the relation of this, if possible, to the level of flow, the type of road, and the economic and climatic characteristics of the region around each site: and (ii) the statistical evaluation, using the results from (i), of optimum methods of counting traffic. Observations were made from March 1968 to July 1970 using Fischer and Porter hourly recording counters at 26 sites, and from September 1968 to November 1970 using SYX-RRL non-recording counters - which were read dafly by observers - at 12 sites. The sites were chosen to be representative of the general range of flow l~vels, road types, and climatic conditions found in Kenya. 3 4. ERRORS IN ESTIMATES OF ADTFROM SAMPLE COUNTS Only where continuous counts are made under perfect conditions can a true ADT or total year’s flow be computed with the expectation of its being absolutely accurate. It follows that any count of less than one-year’s duration must be regarded as a sample, and the estimate of ADT or total years’s flow made from it will be subject to error. The error of estimation is simply the difference between the estimated ADT and the true ADT. If the mean and the standard deviation of these errors are calculated, then probability analysis can be applied to determine, for a given level of confidence, how accurate an estimate of ADT is provided by a particular sample period or sampling procedure. The method of error determination used in the analysis followed the above principles. For a given duration of counting, repeated samples were drawn from the actual flows recorded at each site in one complete year. From each estimated daily flow (ADTE), the true value (ADTT) was subtracted to give the error of estimate. The resulting errors were divided by ADTT and multiplied by 100 to give the proportional error of estimate in percentage terms. This was done so that errors obtained at sites with different flow levels would be on a comparable basis. Thus: proportional error of estimate = 100 (ADTE~~DTT~ percent Finally the standard deviation and the coefficient of variation of the percentage errors were calculated. 4.1 %mple t~ting The errors resulting from the following ADT sampling procedures were determined: 1. Wndom samples of 1,2,3,4, 5, weekdays and 7 consecutive days for au possible periods in the year except those including a Public Holiday. 2. As in 1 for periods of 1,2,4, and 6 whole weeks. 3. Wndom samples of 1,2,3,5 weekdays and 7 consecutive days repeated at regular intervals of three, four, and six months. To provide samples of a reasonable size it was necessary to group the sites by flow level as follows: Group 1 (ADT <75 vehs/day) Group 2 (ADT 76-200 “ ) Group 3 (ADT 201-600 “ ) Group 4 (ADT 601-1000 “ ) Group 5 (ADT >1001 “ ) 4.2 Desirable accuracy of esti mates of traff ic flow To judge the results of the sample tests objectively, it is necessary to decide what level of accuracy estimates of traffic should attain. Specifically we must state within what range of error we wish our estimates of ADT to lie, and how certain we need to be that the estimates lie within the stated range. In the USA3, the accepted standard is that there should be ody a 1 in 20 chance (5 per cent level of probability) that the error of estimate wfll exceed t 10 per cent at any sample count site carrying over 500 vehicles/day. For roads with lower flows, errors of up to t 20 per cent are acceptable. It might be felt that developing countries cannot afford such high standards as the USA, since, the more precise estimates must be, the greater the cost of obtaining them, However, it is considered that accuracy standards in developing countries should be similar to those quoted, and finfact should tend towards the higher of the two, even for roads with low traffic flows. Wereas the use in the USA of lower accuracy standards of traffic counting for roads with low traffic flows is justified to some extent by the relative unimportance of such roads in the USA, the main aim of road improvements in most developing countries is progressf+ely to upgrade earth and gravel roads to bituminous-surfaced roads, i.e. stage construction 4, when the level of traffic demands it. A reasonable standard of traffic estimation is therefore required even for roads with low traffic flows. bwering the confidence limits at which estimates are judged does not seem to be worthwhile since the results rapidly lose any real significance. Until precise studies are completed of the cost-effectiveness of various methods of traffic counting and the sensitivity of the highway planning process to errors in traffic estimates, it will not be possible to specify desirable accuracy limits for developing countries. The USA standards will, however, serve as a criterion by which to judge the performance of estimating procedures elsewhere. 4.3 Practical limits to.sample duration The final point to be discussed before the results are examined is whether there are likely to be any practical limits to the duration of counting in developing countries. In this connection, the most critical consideration is whether counts will be made manually or by machines. Experience of conditions in developing countries su~ests that the great majority of counts will continue to be made manually. The use of automatic counters is at present uncommon and they are only gradually being introduced, mainly for the measurement of seasonal variation and long-term traffic trends. The more widespread use of automatic counters for general counting seems unlikely for some time to come since they are expensive to buy. Nso, they require skflled supervision and maintenance if accurate results are to be obtained, and the necessary skills take time to acquire. bstly, manual methods have the advantage of giving classified counts of traffic flow, and they may dso be politically desirable because of the generally acute unemployment problems. If manual methods of counting are used, then a one-week’s continuous count is about the practicable maximum. Apart from the probable loss of accuracy caused by the boredom of the enumerators, longer counts at each point would reduce the coverage of the road system that was possible. In practice, many counts, although spanning seven days, will probably be for less than 24 hours on some, and possibly all, days. N]@t-time counts are unpopular and difficult to supervise effectively, especia~y in distant ruraI locations. Cost% ffectiveness considerations dso indicate the need to keep the duration of counting as short as possible. Since wages are the main element, the cost of traffic counting can be assumed to increase in direct proportion to its duration. However, simple sampling theory suggests that the accuracy of the resulting ADT estimates is likely to increase in proportion to the sq~are root (approximately) of the duration of counting, i.e. other things being equal, a count for four days will ody double the accuracy of ADT estimation in comparison with that obtained from a singe day’s count, whereas the cost wdl have risen by a factor of four. 5. RESULTS Ml estimate errors given in the results are at the 5 per cent level of confidence. Lower confidence limits 10 per cent (1 in 10 chance) or 20 per cent (1 in 5 chance), can be calculated by multiplying the results by 0.84 or 0.65 respectively. 5.1 Traffic variabili~ and ADT In the early stages of the analysis, it became apparent that the results were stron~y influenced by the ADT at each sample site, so this effect was investigated first. me relationship between traffic variability and flow level is dlustrated in Figure 1 which shows the coefficient of variation of daily flows (~ over a complete year, plotted against ADT. From the Kenya data done, it is clear that the coefficient of variation is inversely related to ADT - the curve has been fitted by inspection as a rough guide. A simple regression analysis based on the relationship, V = ~ (ADT) ‘X ‘ where ~ and ~ are constants, showed that ADT accounts for approximately 55 per cent of the total variation in V. The implication is that, in Kenya at least, the main factor governing the variability of traffic is the average level of flow, and not the function of the road or the type of traffic it carries. Excluding the most extreme of United Kingdom results, Figure 1 shows that traffic variability increases rapidly below flows of approximately 1000 vehicles per day. Partly this is a consequence of the law of small numbers: when the total flow is low a unit change has a proportionately bi~er effect than when the totrd is large. Aso, in practice, variation is inherently greater at low flows because the traffic stream is composed of fewer individual trip motivations, i.e. a flow of 20 vehicles per day on a given road may be motivated entirely by the travel demands of a small government administrative centre, a school, or a sin~e agricultural enterprise. Any change in its activities, such as school holidays, or crop harvesting, can produce very large relative volume changes. Conversely, on roads carrying 500 or more vehicles per day, the trips are usually motivated by a wide range of activities whose operational variations tend to be mutually balancing. men the travel demand for one is high another will be low and vice-versa. Between these two extremes there is a gradual transition and one would expect a steady decrease in variation with increasing flow of traffic. The increase in traffic variability below flows of approximately 1000 vehicles per day is significant because in many developing countries the majority of the rural road system carries daily flows less than this. In Jamaica (1964), Zambia (1 964), and Kenya (1970), the percentages of the rural road system carrying less than 1000 vehicles per day were 95, 98 and 95 respectively 5. Thus in developing countries rural traffic estimation is especially difficult because of the inherent variability of daily travel. &nerally the United Kingdom results do not exhibit any close relationship between traffic variability and flow level. Surprisin~y, over a third of the United Kingdom results are characterised by a very much higher variability than equivalent sites in Kenya. It seems that the effects of climate in the United Kingdom, particularly snow, are much more disruptive than those in Kenya. Aso, because the climatic contrast between winter and summer in the UK is very marked, sites located close to tourist or holiday centres are likely to experience larger relative changes in flow level than equivalent sites in Kenya. Inspection of the locations of sites in the United Kingdom showed that those sites with exceptionally high variability were located either close to tourist-holiday resorts or in areas likely to experience inclement weather. 5.2. Errors from random continuous counts of different durations Table 1 gives the errors for ADT estimates obtained from random continuous counts of different durations. Generally, the errors in estimates fall as both the duration of counting and the ADT increase. There are, however, considerable variations between sites in the rates at which the errors decrease with respect to both the duration of counting and the ADT. &cause of the magnitude of the variations, they are unlikely to be accidental and are probably related to site location and the function of the roads. To make the trends clearer, sites have been grouped into the five flow levels used for repeated sampling and the results averaged. Figure 2 shows that the errors in estimates fall rapidly as the duration of counting and the ADT are increased, but there is a marked decrease in the rate of fall when the duration of counting exceeds a week. This suggests that rural travel is dominated by weekly, rather than by daily or monthly activities. Another significant feature is the sharp fdl in errors in estimates when the duration of counting is extended from five weekdays to one full week. As might be expected, the latter effect is more pronounced at the higher flow levels i.e. on roads that serve regional and district centres with distinctive weekend activities. Clearly, variations in flow at the weekend contribute significantly to total variability, and so if circumstances arise that allow ody a count of 4 or 5 days this period should span tie weekend rather than ordy weekdays. It should be recalled at this point that the suggested maximum practicable duration for a continuous manual count is one week (see 4.3). Figure 2 shows that large errors are associated with counts of only a few days’ duration. The lowest flow-level group has errors in estimates ranging from f 35 per cent for a one-week count to ~ 62 per cent for a count of a singe day. On roads carrying 1000 vehicles per day or less, errors in estimates are at best approximately ~ 20 per cent for a oneweek count and ~ 30 per cent for a one-day count. Clearly then, no practicable duration of random counting is likely to provide ADT estimates of an acceptable accuracy for the great majority of roads in developing countries. 5.3. Repeated random samples Table 2 gives the errors for ADT estimates obtained from repeated random counts of different durations. The figures in brackets show the errors in estimates for continuous counts of equivalent durations. Inspection of the results suggests that the errors in estimates of repeated random counts are related to those obtained from sin~e random counts. If a random count of duration d gives an error equal to *x, then repeating the count will reduce the error to * O= , (n> 1) where n is the number of repetitions (i.e. the errors from repeated counts are approximately proportional to the inverse of the square root of the overall duration of counting). 7 As might be expected, repeated counts give more accurate estimates of ADT than continuous counts of the same duration and the advantage increases with the number of repetitions. Also, repeating a count twice reduces the errors in estimates to approximately 22 per cent of their continuous count value, and repeating four times results in a 40 per cent reduction. However, only at the highest flow levels and for counts repeated four times do the errors in estimates approach the desirable standard oft 10 per cent. Below traffic flows of 600 vehicles per day, repeating counts 3 or 4 times generally results in errors in estimates of between * 10 and f 20 per cent. Because of organisational difficulties, repeated counts are unlikely to be regarded as a practical proposition for most data requirements, although they may be of use for one-off studies. Mso they cannot generally be expected to produce estimates of a desirable accuracy. 5.4 Sampling errors for individual months It seemed likely that random samples drawn from particular months might show errors considerably different from those drawn throughout the year. If a wet season falls consistently in a particular month and normal travel is likely to be interrupted by rain, then samples from that period can be expected to have higher-than-average errors. Conversely, other months, between seasons and away from Public Holidays, could have virtually constant near-average flows, and consequently very low sampling errors. To test this possibility, random samples were drawn separately from each month and the error in ADT estimate calculated as before. The results for one whole week are shown in Table 3. There is considerable variation in errors in estimates from month to month. At any site, the ratio of the largest to the smallest monthly error has a range from approximately 2 to an exceptional 27. Generally, the ratio is in the range 3-12. Even at moderately high flow levels of 400-600 vehicles ADT, the maximum monthly error for one-week random counts can reach 48 per cent. In contrast, even at AD~s of less than 40 vehicles the minimum error in any month does not exceed 10 per cent. Generally, if the month, or months, could be predicted during which sampling errors were likely to be a minimum, then relatively short counts could produce ADT estimates of a high accuracy. Table 4 shows that counts for as few as three consecutive days have errors of 20 per cent or less if the observations are conducted during the month of minimum sampling error. 6. DISCUSSION The magnitude of the errors clearly indicates the need for improvements in the methods of measuring traffic flow in developing countries. In considering how this might be done, it seems useful to examine the basis of methods of counting rural traffic in some developed countries. 6.1 Rural traffic counting in the United Kingdom As mentioned in Section 2.4, rural road planning and design in the United fingdom are based upon the average daily flow (7-day average, 6am-1 Opm) measured in August. Average factors are used to adjust any observations taken in other months to their August equivalent. August conditions are used as the basis for counting because average montNy travel demands have consistently been found to be at their highest then 1. Nthough the method has worked satisfactory in the United Kngdom, because it is based upon a 16-hour observation period it-shares-all the criticisms previously levelled at the other ‘rnhour’ counting systems. Furthermore, although the idea of conducting all counts during a singe period of peak activity is attractive, the method does not seem suitable for use in a developing country, because it presumes that seasonal variations in traffic flow are the same everywhere, and remain so, year after year. Geographical considerations suggest that, in the maidy tropical or sub-tropical developing 8 countries, the climatic variations, and hence most of the likely traffic flow variations, are neither as simple nor as consistent as those experienced in the developed countries. This su~estion is given some confirmation by Figures 3(a) and 3(b) which show the monttiy patterns of flow variation recorded at 21 of the Kenya sites. No simple pattern emerges: the seasonal variation of traffic is higtiy variable both at individud sites and between sites. The causes of the montNy flow variations at each site and whether they recur are stifl being studied, but it is clear that the method of standardizing traffic counts in the United Kingdom could not be used M Kenya, as there is not a sin~e month when flows are near maximum at all sites at the same time. [ 6.2 Rural traffic counting in the USA As explained in Section 2.4, the object of rural traffic counts in the USA is to estimate ADT. men sample counts are made, usually for one or two consecutive days ody, the results are adjusted to give ADT estimates, within the accuracy limits described in 4.2, by factors derived from a relatively small number of continuous counting stations called control stations. This procedure is based upon the fact that seasonal patterns of variation have been found to persist from year to year3 on the same road sections and are simdar: (i) for long consecutive lengths of major road (ADT >500 vehicles) au of which are not necessarily on the same route; and (ii) for dl minor roads (ADT <500 vehicles) within a given economic (geographic) region. ~us the seasonal patterns of traffic variation on all sections of road can be represented by those obtained from either ‘route control stations’ or ‘area control stations’. This system would seem to be the more promising of the two described as far as developing countries are concerned, but it would not be easy to implement. The present sophistication of traffic counting in the USA has been achieved only after many years of recording traffic flows. Wat is apparent is that the use of automatic traffic counters will be a necessary pre-requisite of any significant improvement in the accuracy of current traffic estimates in developing countries. 6.3 Use of automatic traffic counters in developing countries Men first introduced into developing countries, automatic counters should be operated continuously at fixed locations. These should be chosen to represent the major traffic routes and geographic areas; the method of doing this is described in a recent Reports. As well as monitoring long-term trends, the counter results will enable a study to be made of the magnitude, frequency, and causes of the day-to-day and month-to-month, fluctuations in flow. A clear understanding of these will enable methods of counting traffic to be designed along the lines indicated, so that ADT estimates of a prescribed accuracy can be made. After one or two years, additional counters could be obtained and a start made on the grouping of road sections according to their seasonal variation characteristics. In the USA seasonal variation ‘ counts are made for ody one week in every month at a given location; with efficient organisation, a single counter can therefore cover four sites per year. In the initial stages of such a system, there is no need for expensive makes of traffic counter to be used. The simple SYX-RRL Nos. 4A or 4B accumulating counter would be adequate and it costs approximately one-tenth as much as a recording counter. A recent report by the Road Research Laboratory describes the operation and maintenance of the SYX-RRL counters under tropical conditions. 9 7. CONCLUSIONS 1. Traffic counts in developing countries should seek to provide estimates of the Annual Average Daily Traffic (ADT) on a road. 2. If made manually neither simple random traffic counts nor replicated random counts of any practicable duration can provide estimates of ADT within desirable limits on the majority of roads in developing countries . 3. Any appreciable improvement in estimates of traffic flow in developing countries will require the use of automatic traffic counters operated continuously at fixed locations. 8. REFERENCES 1. DUNN, J.B. Traffic census results for 1969. Department of the Environment, RRL Report LR371. Crowthorne, 1970 (Road Research bboratory). 2. MINISTRY OF TRANSPORT. Layout of roads in rural areas. London, 1968 (HM Stationery Office). 3. US DEPARTMENT OF COMMERCE. Guide for traffic volume counting manual. Washington DC, 1965, (Bureau of Public Roads). 4. UNESCO. hw cost roads: design, construction and maintenance. Drafted by L. ODIER, R. S.MILLARD, PIMENTAL dos SANTOS, S.R. MEHRA, hndon, 1971 (Butterworths). 5. HOWE, J. D.G.F. Kenya 6@point traffic census: Design and results for 1970. Department of the Environment, RRL Report LR 398. Crowthome, 1971 (Road Research bboratory). 6. BLACKMORE, D.H. and J. D.G.F. HOWE. Using SYX-RRL vehicle counters numbers 4A and 4B in tropical countries. Department of the Environment, RRL Report LR 385. Crowthorne, 1971 (Road Research bboratory). 10 TABLE 1 Errors in ADT estimates from random ADT ~vehicles) 25 26 26 32 44 56 63 88 93 106 120 152 156 250 355 357 438 494 501 counts of vawing duration (per cent ) Number of Weekdays 1 2 3 4 5 70.6 58.8 50.2 48.0 41.4 69.0 57.0 55.1 51.7 53.1 80.6 69.4 63.7 61.2 58.8 83.9 71.7 66.2 66.6 65.8 41.7 31.8 27.2 26.8 25.1 37.4 30.4 27.2 27.4 25.5 52.9 43.7 40.8 37.0 35.5 44.1 40.8 39.8 34.9 30.6 51.0 45.3 42.9 ,40.8 43.7 45.7 38.6 34.7 31.8 28.4 41.6 34.3 25.5 25.9 29.4 39.0 33.9 36.3 31.4 27.0 36.8 32.1 30.0 28.4 25.5 50.8 47.0 44.5 43.5 41.4 35.7 28.4 27.2 26.8 30.2 41.7 37.2 32.7 30.8 30.4 37.2 36.1 34.9 36.8 34.9 38.4 30.8 27.6 25.5 25.1 36.8 33.7 33.1 32.9 31.8 Number of Weeks 1 2 4 6 36.3 34.1 33.9 34.9 31.2 28.4 23.3 19.4 50.2 44.7 38.0 33.1 54.7 55.7 60.0 63.3 22.5 19.2 16.8 17.4 22.9 19.4 14.9 9.2 30.0 28.4 26.8 23.3 26.5 22.5 20.4 16.1 33.1 32.9 31.6 30.2 24.1 20.8 19.0 18.8 22.3 20.2 17.4 14.5 34.5 30.6 22.5 7.4 18.8 14.1 10.1 6.3 39.8 34.7 34.5 32.5 13.1 12.3 11.8 10.2 36.6 32.7 19.0 8.2 22.1 17.6 14.1 5.5 18.0 17.4 17.8 16.7 29.4 27.6 21.6 16.5 TABLE 1 (continued) ADT ‘veticles) 1 2 3 4 5 1 2 4 6 622 25.3 20.4 17.2 17.2 17.2 17.4 13.9 9.6 7.8 632 40.0 35.3 32.9 29.4 27.4 24.9 23.3 22.1 21.4 650 32.3 34.3 36.4 34.5 32.7 22.7 17.2 9.8 7.0 675 25.7 23.9 23.5 22.0 20.2 21.2 17.4 11.2 7.8 676 30.4 31.8 34.3 31.8 32.5 20.8 19.0 16.8 16.7 788 26.6 20.6 19.8 19.2 16.3 19.0 16.5 14.9 14.5 792 39.7+, 44.1 45.7 45.9 41.0 24.5 22.3 16.3 16.3 . 825 20.8 19.0 18.6 16.8 17.6 19.0 16.8 13.7 12.7 1109 24.5 21.8 19.4 17.8 15.3 16.1 14.3 14.7 14.5 1185 31.4 30.2 30.4 30.2 28.6 18.2 17.0 17.0 17.2 1250 35.3 28.0 26.7 22.0 21.2 15.3 13.5 11.0 9.6 1373 34;3 40.4 43.1 41.2 36.1 16.5 13.7 9.8 8.8 1751 30.0 27.2 26.5 25.1 22.2 18.0 17.0 17.8 17.4 1766 26.8 29.8 31.2 31.4 26.1 12.9 11.2 8.6 7.1 2846 31.2 29.0 28.4 29.0 27.0 16.3 12.7 7.8 6.9 TABLE 2 Errors in ADT estimates from repeated random counts of varying duration (per cent) Duration of counting flow level (vehicles/day) Repetitions Number of weekdays 1 week 1 2 3 5 <75 1 62.3 51.7 47.2 43.5 35.5 2 46.6 (5 1.7) 36.4 31.4 .~8.8 23.5 (32.9) 3 33.5 (47:2) 28.0 25.5 23.5 17.6 4 25.9 (45.5) 24.3 22.1 20.4 16.1 (30.6) 1 43..1 37.4 34.9 30.8 26.6 2 28.0 (37.4) 22.9 24.5 21.2 17.6 (23.5) 75-200 3 23.7 (34.9) 18.6 17.4 15.7 14.3 4 18.6 (32.1) 17.2 15.9 13.9 12.5 (20.2) 1 40.2 35.5 33.3 32.3 26.5 2 32.7 (35.5) 27.8 25.1 21.6 17.6 (23.7) 201-600 3 22.7 (33.3) 22.3 18.0 17.4 14.3 4 21.8 (32.7) 20.2 15.7 15.1 12.5 (19.8) 1 30.0 28.6 28.6 25.7 21.2 2 19.2 (28.6) 15.5 17.6 17.0 13.9 (18.4) 601-1000 3 16.3 (28.6) 15.5 15.9 13.9 11.4 4 14.3 (27.0) 11.2 12.3 12.2 10.0(14.3) >1000 1 30.6 29.4 29.4 25.3 16.3 2 17.8 (29.4) 18.4 16.8 14.7 13.3 (14.3) 3 15.7 (29.4) 15.7 14.9 12.9 11.8 4 14.3 (29.0) 11.0 11.8 11.2 7.6 (12.3) Figures in brackets are for continuous counts of an equivalent duration. ADT ~ehicles) 25 26 32 44 88 93 106 156 250 349 357 438 494 501 622 632 675 676 788 792 825 1109 1185 1250 1751 2846 TABLE 3 Errors in ADT estimates from random counts of 1 week in different months (per cent) . Jan Feb Mar Apr May Jun Jul Aug Sep Ott Nov Dec 48.8 31.8 13.9 10.8 6.9 10.0 23.9 18.2 11.0 6.7 9.2 10.8 6.7 13.7 9.0 8.6 17.6 7.6 15.5 6.1 9.0 8.6 3.9 11.0 17.2 8.0 15.9 22.s 40.6 4.9 12.0 8.8 22.5 6.7 8.4 9.6 5.3 4.7 9.0 3.3 3.9 6.9 10.8 4.1 2.4 10.2 8.6 2.0 8.6 5.3 10.8 3.7 27.4 8.4 39.8 16.1 25.3 12.9 34.3 10.0 8.0 9.6 8.2 13.9 28.6 8.6 38.8 8.0 6.1 6.3 6.3 9.2 7.4 8.4 8.2 4.7 4.3 11.0 13.3 23.1 28.6 15.9 14.7 5.5 12.3 31.4 17.4 5.9 11.0 5.3 12.5 19.8 4.3 13.5 5.5 3.7 3.7 5.3 7.4 3.3 4.3 11.0 3.1 2.9 10.0 33.3 23.3 11.4 36.1 18.8 17.8 28.0 8.6 5.5 78.0 9.4 8.4 ‘11.2 7.4 5.1 6.3 8.0 9.2 11.8 9.0 4.5 7.8 12.7 11.4 11.0 27.0 18.0 11.8 21.8 11.8 6.1 6.3 5.5 18.8 7.0 4.3 4.7 6.9 5.1 12.2 21.2 14.7 2.4 6.7 6.1 3.7 3.3 8.6 7.6 4.9 7.8 7.0 64.3 9.6 9.8 11.0 9.0 7.6 12.0 9.2 5.5 9.2 5.1 6.5 8.8 4.3 8.4 8.0 17.2 12.9 21.8 6.9 7.0 7.2 8.6 9.0 7.6 23.9 22.3 14.1 12.0 21.6 10.8 16.3 6.3 5.7 7.4 23.5 3.5 7.6 11.6 15.9 18.4 3.7 17.4 6.9 8.0 5.5 18.0 8.2 7.8 5.9 13.9 21.4 25.1 14.5 13.1 14.1 31.0 14.5 20.2 18.0 9.2 4.3 10.0 6.9 9.2 10.6 25.5 13.1 9.0 5.9 11.0 6.5 8.2 17.2 14.3 8.8 12.5 17.4 38.4 14.1 7.0 15.5 26.6 20.8 5.9 19.6 8.2 2.9 10.4 16.5 7.4 20.0 25.9 14.5 4.3 6.5 35.5 10.2 4.7 6.1 6.5 10.0 22.7 27.2 34.7 18.8 18.4 17.6 20.8 30.0 11.8 45.5 9.2 10.2 7.8 9.8 10.4 17.6 7.0 5.7 19.0 8.2 7.6 8.4 7.0 26.6 5.5 5.9 26.3 26.1 35.5 12.9 5.7 23.9. 45.7 15.3 9.8 18.4 8.0 17.6 28.6 14.1 48.0 6.7 9.6 15.9 15.3 30.4 8.4 20.0 22.0 10.0 15.9 19.2 6.3 Yearly Average 36.3 50.2 54.7 21.8 26.5 33.1 24.1 18.8 39.8 13.1 36.6 22.1 18.0 29.4 17.4 24.9 16.7 20.8 19.0 24.5 19.0 16.1 18.2 15.3 18.0 16.3 -,, - ,.. ., .,,,“ ,. ,.,., “, h“. ,., . ,, ADT (veticlea) 25 26 32 44 88 93 106 156 250 349 357 438 494 501 622 632 675 676 788 792 825 1109 1185 1250 1751 2846 TABLE 4 Errors in ADT estimates from random counts of 3 consecutive weekdays in different montk (per cent) Jan Feb Mar Apr May Jun Jd Aug Sep &t Nov Dec 82.3 42.7 31.8 17.2 13.9 13.5 22.0 21.0 28.8 8.6 12.3 18.8 7.6 11.8 12.3 10.8 20.4 4.9 15.1 9.0 15.3 6.9 7.2 9.8 19.0 20.0 17.0 37.6 54.1 10.2 25.1 16.7 32.1 6.7 25.1 14.5 9.2 13.5 13.3 5.9 10.0 12.5 7.8 4.5 5.5 12.3 8.4 6.3 8.2 8.6 13.7 6.3 33.3 17.4 71.0 20.2 37.2 17.6 51.5 9.8 11.4 13.3 13.1 16.1 29.4 10.2 28.4 11.0 6.1 9.2 6.7 9.6 12.3 12.2 4.9 5.3 5.1 13.3 23.1 31.0 52.7 18.8 33.5 15.1 17.2 54.1 14.9 8.8 33.5 6.9 11.0 23.5 9.6 19.2 18.2 6.7 12.0 13.9 9.8 10.2 12.9 31.8 17.0 10.2 19.2 54.5 31.0 14.3 54.9 30.6 30.6 42.3 19.6 9.0 69.6 7.8 7.8 14.7 16.8 6.3 10.6 8.2 18.0 11.4 8.6 4.3 6.9 20.2 10.4 8.6 21.2 36.1 22.7 32.9 23.1 13.1 16.7 12.0 30.2 11.2 4.9 6.1 12.9 5.9 6.5 23.3 22.9 3.9 12.9 10.0 6.9 9.6 7.8 11.2 8.2 13.1 21.4 39.6 103.7 38.4 20.0 21.4 14.9 19.0 15.3 28.0 14.7 14.1 16.1 18.8 19.4 14.1 13.3 10.6 10.2 14.3 11.8 31.4 7.4 8.0 13.9 14.5 12.7 18.4 8.6 24.9 10.8 18.8 13.7 5.1 22.7 31.0 13.5 7.8 25.9 11.8 10.4 5.3 8.2 28.2 10.6 14.1 11.4 9.4 14.9 9.0 10.4 9.8 22.5 23.5 31.4 23.5 19.2 33.3 16.7 19.6 25.5 13.7 6.7 14.1 12.7 15.7 15.1 29.4 18.8 8.2 7.2 12.2 13.1 10.0 18.8 15.7 11.4 14.3 17.4 38.6 33.1 19.0 42.7 41.9 17.8 12.9 25.1 15.9 6.7 20.6 24.1 12.9 14.5 38.6 16.5 4.3 18.6 35.5 12.2 10.8 7.0 11.4 16.3 22.3 29.0 35.1 31.9 34.5 20.6 35.9 40.2 17.4 65.8 13.5 20.4 11.4 18.8 8.2 20.8 13.7 8.2 20.0 11.6 9.2 10.4 7.4 25.5 7.4 6.3 31.8 64.7 38.0 30.0 18.6 19.0 49.2 29.0 10.0 42.9 15.3 17.0 53.9 32.9 53.7 17.2 15.3 25.7 29.8 28.8 22.0 11.4 32.9 32.7 35.9 28.4 16.5 Yearly Average 50.2 63.7 66.2 27.2 40.2 42.9 34.7 30.0 44.5 27.2 32.7 34.9 27.6 33.1 17.2 32.9 25.1 34.3 19.8 45.9 17.6 19.4 30.4 26.6 26.5 28.4 COUNTRY COLOMBIA CYPRUS ETHOPIA 8. APPENDIX I . Questionnaire on national procedures for sumey of rural traffic flow Is there any form of national traffic census operating? Yes Yes Yes NATIONAL CENSUSES For how many hours and days are observations made duringa survey? 24 hours a day ‘or 7 days. 7 days continuous y on some trunk “oads;92 hours ;pread over 5 day! )n others. 84 lours spread over 5 days on tourist and village roads, but varies from year to year. 24 hours a day for 5 days. How many surveys are made a year? 1 3 3 Jan-April dry season; June-August wet season; Sept.-Dee. intermediate season. Are all classes of road covered? kly roads mainained by national kvernment about 7 sites ler year). Yes Al-weather road built and maintained by the Imperial Highwa] Authority whether primary, secondary or feeder roads. GENERAL COUNTS For how many Lours and days [re observations made? 4 hours a day )r 7 days mainly. /aries 24 hours for 7 to 15 days. Is there any specific period, or periods, of the year when observations are made? Vo. Varied from )lace to place :0 coincide with ocal croplarvesting period No Does the ;overnment se standards for traffic observations? ‘es. ‘es 7es AUTOMATIC COUNTERS Are automatic traffic counters in use? ‘es. Jo Vo For what purpose? b continuous :ounters at ixed ocat ions. APPENDIX I - continued (1) NATIONAL CENSUSES T For how many How many hours and days surveys are ire observations made a year? made during a survey? I GENERAL COUNTS AUTOMATIC COUNTERS For what purpose? As continuous counters at fried locations Are afl classes I For how many of road, hours and days Is there any pecific period, or periods, of :he year when observations are made? Does the ;overnment set standards for traffic observations? {0. Are automatic .raffic counters in use? Yes. ;OUNTRY Is there any form of national traffic census operating? covered?- I areobservations made? GHANA Yes 7 days 12 hours (6 a.m.-6p.m.) manually and 24 hours by automatic counter. 3 or 4 times a year. No. No (one is proposed for 1971) ~ days Up to 2 times a year in each season 40. To a limited extent. [RAN 1st day: midnight-8 a.m. 2nd day: 8 a.m.~ p.m. 3rd day: 4 p.m.-midnight. Yes 12 hours (6 a.m.- 6 p.m.) for 7 days. Wet weather anc main annual social events avoided. RENYA Yes 12 hours (6 a.m.- (es No 6 p.m.) for 5 days in a week and 24 hours on the remaining 2 days. 4 February, May, August November. LESOTHO Yes 12 hours a day Irregular No for 7 days No Yes For short count of one day’s duration. MALAM Yes 1 day 12 hours manual and 24 hours automatic counter. 2 Wet and dry season. Main and secondary roads only. APPENDIX 1- continued (2) NATIONAL CENSUSES GENERAL COUNTS , AUTOMATIC COUNTERS COUNTRY IS there any form of national traffic census operating? For how many hours and days are observations made during a suNey? How many surveys are made a year? Are all classes of road covered? For how many hours and days are observations made? Is there any pecific period, or periods, of he year when observations are made? Does the 2overnment set standards for traffic observations? Are automatic traffic counters in use? For what purpose? No 12 hours (6 a.m.- j p.m.) for 7 days. MAUWTIUS kring sugar :rop period Jul} o December. TANZANIA Yes 72 hours 2 Yes 72 hours Yes New standards being introduced. All roads to be I day 8 hours nanual (8 a.m.- 1 p.m.) and 24 hours by automatic counter. Yes ‘n the past for ihort counts of )ne day’s duration but permanent sites are to be set up. covered. Most counts 2 times a year, April and October, 8-hour counts (8 a.m.4 p.m.) for 5 weekdays. At a few sites 4 times a year January, April, July, October, 8-hour counts spanning 24 hours for a total duration of 3 weeks; i.e. an effective oneweek count. 8-hour counts Ordy those usually 3 times which are the a year but responsibility some just once of the national or twice 24- highways hour counts department. repeated up to 15 times a year. 2 Only roads which are the responsibtiity of the Ministry of Works. Yes Variable, majority are l-day 8-hour counti. Some are for 24 hours 1 or 2 days. Only in urban areas. UGANDA 24 hours a day for Yes 7 days. Yes For short counts of 7 days duration. 24 hours a day for 7 days. No Yes APPENDIX I - continued (3) COUNTRY ZAMBIA Is there any form of national traffic census operating? Yes NATIONAL CENSUSES For how many hours and days are observations made during a survey? 12 hours for 5 days in a week and 24 hours on the remaining 2 days. How many surveys are made a year? Once a year in June. Are all classes of road covered? Yes GENERAL COUNTS For how many hours and days are observations made? Varies from 12 hours up to 7 days. IS there any specific period, or periods, of the year when observations are made? D~ season MayOctober Does the Government set standards for traffic observations? Yes AUTOMATIC COUNTERS Are automatic traffic counters in use? To a limited extent. For what purpose? For short ;ounts of 30 iays duration. ,,,—-—— m 70 60 50 40 30 20 10 0 D u 0 0 ● Kenya O United Kingdom A Nigeria o 0 0 0 0 ● D ● o b o ‘~,” w ● n A A w ● A o 500 1000 1500 2000 2500 3000 AD T- vehicles Fig .1. RELATIONSHIP BETWEEN OAILY VARIABILITY ANO flOW LEVEL \ i <75 vehicles/day 201- 600vehic les/day 76-200 vehicles/day ---- _____ >1001 vehicles/daY I 1week 2 weeks 4 weeks 6 weeks 1 1 1 1 12345 10 15 20 25 30 35 40 45 Weekdays Duration of counting (days) Fig.2. ERRORS IN ADT ESTIMATES FROM RANOOM COUNTS OF VARVIN6 MRATION la r ADT 26 1 140~ ADT 63 1 120 Per cent 100 80 w ,0 ~ JFMAMJJASOND 140 r ADT 39 1 120 Per cent 100 80 w 60 ~ JFMAMJJASOND 140 ADT 41 120 Per cent 100 60 60 M JFMAMJJASOND 140r ADT 56 1 120 Per cent 100 80 H 60 ~ JFMAMJJASOND 140 ADT 60 120 Per cent 100 80 K 60 ~ JFMAMJJASOND 120 Per cent 100 80 H 60 ~ JFMAMJJASOND 140 120 Percent 100 80 60 r ADT 80 ~ k IF MA MJJASOND 140 [ ADT 93 120 1 ‘ercent’00+ :U JFMAMJJASOND 140 120 Per cent 100 80 60 ADT 95 P 111111111111 JFMAMJJASOND 140r ADT 120 1 Per cent ~w 80 t i 60 ~ JFMAMJJASOND Fig.3(a) MONTH-TO-MONTH VARIATIONSINAVERAGE TRAFFICFLOW ON WEEKOAVS-KENVA (AOT 26-120) 140 r ADT 152 1 140 r ADT 650 1 120 Per cent 100 80 H 60 ~ JFMAMJJASOND 140 r ADT 152 1 120 Percent 100 80 60 w JFMAMJJASONO !40 ADT 344 120 Percent 100 80 w 140 r ADT 418 1 120 Percent 100 80 w Go ~ JFMAMJJASOND 140 r ADT 619 1 Per cent ~w :M JFMAMJJASOND lLO 120 [ Per cent ‘“M :U JFMAMJJASOND 140 r ADT 683 1 120 Percent 100 80 w 60LII I 1 i I I 1 I 1 I JFMAMJJA SONO 140 r ADT 717 1 Per cent 100 80 w 140r ADT 782 1 Percent 100 80 60 m JFMAMJJASOND ?40 r ADT 1373 1 Per cent ~M ;UJFMAMJJASOND ADT 1766 1 Per cent ‘oo& 80 60 1 I 1 I I I 1 1 I I JFMAMJJASOND Fig. 3 (b) MONTHTO MONTH VARIATIONSINAVERAGE TRAFFICFLOW ON WEEKOAV - KENVA (AOT152-1766) (1718) Dd6S3271 4M 12/71 HtP G191S PRINTED IN ENGLAND ABSTRACT Areviewof rural traffic-counting methods in developing countries: JDGFHOWE, MSCB Tech: Department of the Environment, RRL Report LR 427: Crowthorne, 1972 (Road Research Laboratory). This Report reviews methods of rural traffic counting currently used in developing countries and examines the accuracy of the resulting flow estimates. The results of a questionnaire survey among a sample of developing countries suggest that decisions on the duration, frequency, and timing of counts are at present arbitrary. Consequently, estimated daily traffic flows can rarely be expected to lie within *3O per cent of the true value averaged over the whole year. Although repeating counts at intervals throughout the year increases the accuracy of traffic estimates, this is achieved only at a disproportionate increase in cost. It is concluded that for any appreciable increase in the accuracy of rural traffic estimates much more needs to be known about the magnitude and causes of the variations in flow. This requires that automatic traffic counters be used on a wider sca”le than at present. ABSTRACT A review of rural traffic-counting methods in developing countries: J D G F HOWE, MSC B Tech: Department of the Environment, RRL Report LR 427: Crowthorne, 1972 (Road Research Laboratory). This Report reviews methods of rural traffic counting currently used in developing countries and examines the accuracy of the resulting flow estimates. The results of a questionnaire survey among a sample of developing countries suggest that decisions on the duration, frequency, and timing of counts are at present, arbitrary. Consequently, estimated daily traffic flows can rarely be expected to lie within *3O per cent of the true value averaged over the whole year. Although repeating counts at intervals throughout the year increases the accuracy of traffic estimates, this is achieved only at a disproportionate increase in cost. It is concluded that for any appreciable increase in the accuracy of rural traffic estimates much more needs to be known about the magnitude and causes of the variations in flow. This requires that automatic traffic counters be used on a wider scale than at present.