- 最后登录
- 2014-10-14
- 注册时间
- 2014-8-2
- 威望
- 0
- 金钱
- 105
- 贡献
- 35
- 阅读权限
- 20
- 积分
- 140
- 日志
- 0
- 记录
- 0
- 帖子
- 15
- 主题
- 5
- 精华
- 0
- 好友
- 0
  
- 注册时间
- 2014-8-2
- 最后登录
- 2014-10-14
- 积分
- 140
- 精华
- 0
- 主题
- 5
- 帖子
- 15
|
罗老师,您好!之前向你请教了icc相关问题,获益良多。随后进行了跑数据活动,发现一个问题,向你请教请教。我建立的模型使用两个指标作为因变量。一个是roa,一个是eps。但我发现在,roa的离异数(Deviance)出现了负值,不论是零模型还是随机系数模型亦或完整模型。6 y5 c G: W/ K4 k- k
我查询了一些文献:4 I9 L3 \ }* Z
同一研究可以使用多个HLM模型,由此产生一个问题,多个HLM模型的优劣如何判断?如果两个模型是嵌套模型,可以通过离异数(-2LogLikelihood,-2LL)来判断,由于-2LL值近似服从χ2分布,如果两个模型的-2LL值之差的χ2检验显著,表明改进的模型更优。——我国近十年来心理学研究中HLM方法的应用述评 方杰 邱皓政 张敏强 方路4 Y! S0 j* L H. x+ i
Importantly,PROCMIXED produces a variety of statistics useful for comparing the goodness-of-fit: n( D8 [; G8 p* ~" f
ofmultiplemodels using the degrees of freedom(df) associated with the number ofmodel differences
9 f6 F' ^, [5 N" k/ l, Qbetween the contrasted models. Of particular relevance are the 2 log likelihood. V5 o7 L$ Q/ l$ q
(2LL, also referred to as the deviance statistic) and Akaike’s information criterion (AIC).& w4 a6 S- ?2 E: S8 b3 `
In both cases,models that fit better produce smaller values. The2LL depicts difference scores
; b+ U+ W& M# w* R& M/ H" Tbetween nested models—for example, comparing one model containing n predictors with- {! B+ ^4 K4 N# Y3 _
another containing n1 predictors—and follows an approximate w2 distribution. Thus, unlike
: c+ b& \9 z& M( B7 I! ]3 pAIC, it offers a w2 significance test. For model comparison purposes and for consistency with
$ R( e3 q- n' oprevious studies, we report the 2LL and evaluate goodness-of-fit using the w2 test.
5 T* v4 W/ \5 `4 t- ]( b5 Z: z- Q4 h3 |$ @' f( [) R" _
Modeling Levels and Time in Entrepreneurship Research An Illustration With Growth Strategies and Post-IPO Performance2 x! S7 h4 f6 X5 O
Tim R. Holcomb James G. Combs David G. Sirmon Jennifer Sexton; O$ M/ @4 L7 n' a2 U# c! Y
+ c* d5 V! g$ X; {) D% S
文献中提到了多个模型的比较,我从温福星老师的书中也发现通过模型的离异数来计算模型优劣程度。并没有提及假如离异数为负值,是哪些方面出了问题?是否在数据方面有异常? |
|