Suchergebnisse

In this paper, the efficiency of autoregressive integrated moving average (ARIMA) and regression models in simulating air-transport passengers by route are compared and constrasted. It is concluded that ARIMA models are far superior not only in their simulation capabilities but also in their applicability to such data. In the context of the UK Civil Aviation Authority's approach to forecasting, it is suggested that ARIMA models, including those with intervention terms, bear closer examination.

This paper reviews the comments made by Gordon (1979) on the paper by Pitfield (1978) and, where appropriate, provides further evidence. In particular, intrazonal error in the gravity model is examined for twenty-eight commodities and it is concluded that the error is not as serious as implied. There is no evidence for bias in the diagonal elements of the distance matrix, and a deterrence function, used as an alternative to the negative-power function, does not provide better predictions. It is also shown that the PEP statistic is not corrected for scale in the same way as the other standardised statistics and so its correlation with these other statistics is strongly influenced by the scale of entries in the matrices. Minor points are made on the questions of speed of gravity-model convergence as β → ∞ and on the suitability of the χ2 test.

The transportation problem and a doubly constrained gravity model with a power deterrence function are used to find predictions of a number of 134 × 134 freight matrices detailing tonnages moved in Great Britain in 1972. The matrices detail movements by thirty commodity groups, and predictions are obtained for movements by road for all but one of the commodities and for the principal items carried by rail. These predicted matrices are used to examine a number of questions. The relationships between some alternative goodness-of-fit statistics are examined to establish which commodities are best modelled by each technique and to point out empirically which statistics give unreliable rankings. Various summary measures of the actual matrices are examined to see if it is possible to predict which matrices will be best modelled by each technique. The modelling techniques are compared to indicate which provides the best predictions for each matrix, and some conclusions are offered on the absolute efficiency of the best models.

Covariance analysis is used to explore two questions encountered in modelling freight generation for a multizonal system. At the level of the British standard region, where employment data by industry are readily available, experiments are conducted on a variety of aggregations of freight-flow data to see whether explanatory-variable data collection and manipulation can be minimized. This would be the case if common relationships between freight generated and employment data hold for broad groupings such as 'heavy' industry, chemicals, and engineering. At the zonal level an examination is made of the spatial distribution of residuals from the regression of total road freight generated on resident population. It is shown that for different zone types different relationships hold, and it is demonstrated that the centre–periphery explanation of the spatial distribution of residuals can be seen to be valid if 'centre' and 'periphery' are defined in a functional sense.