There is a strong evidence of over-

There is a strong evidence of over-dispersion for all models, which is indicated by the significant estimated
dispersion value in the Negative Binomial regression as shown in Table 1: 0.811, 0.667 for vehicle disabilities
and crashes in environmental effect analysis, and 0.673, 0.770 for two types of incidents in traffic effect analysis
respectively.
Since the number of observations for each day is 12 and for each segment is also 12, the correlation structure is
a symmetric matrix and its dimension is 12 with one in each diagonal position. The autoregression structure
assumes that the correlations between the multiple observations for a certain day will decrease as the time-gap
increase and the correlation is multiplied by the parameter for one more hour gap. In this case, the correlation
estimated is 0.223 in environmental analysis models for vehicle disabilities. That is to say, the autoregression
structure has correlation of 0.223 for each successive two hours and correlation 0.050 for time periods with onehour
gap. This conclusion in turn has proved the rationality of choosing weather condition in previous two hours
in Pre. Snow and Pre. Rain variables, since the correlation could be ignored for hours with two-hour gap or
more. The correlation parameter is 0.240 for crashes, and a similar conclusion could be drawn. In traffic effect
analysis models, the correlation parameter values are 0.705, 0.770 for vehicle disabilities and crashes, much
higher than those in environmental analysis. This may be because the correlation of traffic condition among
different hours is more closely. As a result, the effect of traffic condition on incident occurrence could last 4
hours or more (the correlation of time periods with three-hour gap is about 0.25).
Moreover, it is found that, both in environmental and traffic effects assessment, the estimated correlation
parameters of vehicle disabilities are lower than those of crashes. This indicates the occurrence of crashes is more
correlated in a temporal scale.

結果 (繁體中文) 1:

強有力的證據證明過度分散體為所有模型，這是由大量的估計在負二項式回歸如圖所示表 1: 0.811，車輛殘疾 0.667 色散值和環境影響分析，和 0.673，0.770 為兩種類型的事件在交通影響分析的崩潰分別。因為對每一天的觀察次數 12，並為每一段也是 12，相關結構是對稱矩陣和它的尺寸是 12 與一個在每個對角線的位置。自回歸結構假設某一天多的觀察所得的相關性將隨時間差距增加和相關性被乘以一個多小時差距的參數。在這種情況下，相關估計是 0.223 車輛殘疾的環境分析模型中。即是說，自回歸結構具有關聯的每個連續的兩個小時為 0.223 和相關 0.050 為 onehour 的時間段差距。這一結論反過來證明在前兩個小時內選擇天氣條件的合理性在前。雪和前。雨變數，因為好幾個小時，兩個小時的差距可以忽略相關性或更多。相關參數是 0.240 的崩潰，和可以得出類似的結論。在交通影響分析模型的相關參數值是 0.705、 0.770 為車輛殘疾和崩潰，很多higher than those in environmental analysis. This may be because the correlation of traffic condition amongdifferent hours is more closely. As a result, the effect of traffic condition on incident occurrence could last 4hours or more (the correlation of time periods with three-hour gap is about 0.25).Moreover, it is found that, both in environmental and traffic effects assessment, the estimated correlationparameters of vehicle disabilities are lower than those of crashes. This indicates the occurrence of crashes is morecorrelated in a temporal scale.

結果 (繁體中文) 2:

有很強的證據過度分散對所有模型，其由顯著估計表明
在負二項回歸色散值如表1所示：0.811，0.667車輛殘疾
和崩潰在環境效應的分析，和0.673，0.770對於兩種類型的事件中的流量影響分析的
分別。
由於觀察結果為每個天數為12，並為每個段也是12，相關結構是
一個對稱矩陣和它的尺寸是12與一個在每個對角的位置。自回歸結構
假定多次觀測一定天之間的相關性將隨著時間間隔
的增加，相關被乘以1小時以上的間隙的參數。在這種情況下，相關
估計是在對車輛殘疾環境分析模型0.223。也就是說，自回歸
結構具有對於每個連續的兩小時，相關0.050與1小時時間段相關的0.223
間隙。這一結論又證明了選擇在以前2小時氣象條件的合理性
在前期。雪和Pre。雨的變量，由於相關可為小時忽略了兩個小時的間隙或
更多。相關參數是0.240的崩潰，和類似的結論可以得出。在交通影響
分析模型，相關參數值是0.705，0.770車輛障礙和崩潰，遠
高於環境分析更高。這可能是因為交通狀況之間的相關性
的不同小時是更緊密。其結果是，交通條件對入射發生的效果可能會持續4
小時或更長時間（在三個小時的間隔時間段的相關性為約0.25）。
此外，人們發現，無論是在環境及交通的影響的評估，該估計相關
車輛殘疾參數比崩潰低。這表示崩潰的發生是更
在一個時間尺度相關。

結果 (繁體中文) 3:

有一個强有力的證據表明，在負二項回歸的負二項回歸的顯著的估計，這是在錶0.811:1，0.667的車輛殘疾，0.673，0.770，，2種類型的事件在交通影響分析的顯著的過度分散。由於每天觀測的觀測值為12，並且每一段的觀測值也為12，相關結構為對稱矩陣，其維數為12。自回歸結構
假定某一天多個觀測值之間的相關性會隨著時間
减少差距新增和相關的參數為一個多小時的差距乘以。在這種情況下，相關估計為0.223，在環境分析模型的車輛殘疾。也就是說，自回歸
結構具有時間
間隙時間週期的每個連續的兩小時0.050 0.223相關關係。這一結論證明了前2個小時內選擇天氣條件的合理性。雪和前。雨變數自相關可以忽略幾個小時或兩小時的差距
更多。的相關參數為0.240的崩潰，和一個類似的結論可以得出。在交通影響分析模型中，相關參數值分別為0.705、0。