where is the attachment??

<p>data a;<br/>input subject&nbsp;anxiety&nbsp;tension&nbsp;trial&nbsp;score@@;<br/>cards;<br/>/*数据粘贴*/&nbsp;<br/>run;<br/>proc glm data=a;<br/>class　subject&nbsp;anxiety&nbsp;tension&nbsp;trial;<br/>model score=subject&nbsp;anxiety&nbsp;tension&nbsp;trial / ss3;<br/>means subject&nbsp;anxiety&nbsp;tension&nbsp;trial/ regwq;<br/>run;</p><p><br/>结果是这样的组合得分最高，对不起，我也不会~~~</p><p>subject=10<br/>anxiety=2&nbsp;<br/>tension=2&nbsp;<br/>trial=1<br/></p>

[此贴子已经被作者于2007-11-25 23:47:46编辑过]

<p>data scoredata;<br/>input subject&nbsp;anxiety&nbsp;tension&nbsp;trial&nbsp;score@@;<br/>cards;<br/>/*数据粘贴*/&nbsp;<br/>run;<br/></p><p>**the default method of estimate the prarameter&nbsp;is REML, you can use method=ml to change it maximum likelihood**;</p><p><br/>proc mixed data=scoredata method=ml;<br/>class subject anxiety tension;<br/>model score= trial anxiety tension/S;<br/>random int;<br/>run; </p><p><br/>proc mixed data=scoredata method=ml; <br/>class subject anxiety tension; <br/>model score = trial anxiety tension/S; <br/>repeated / type=cs subject=subject; <br/>run;<br/></p>

<p>Model 1 is called random intercept model. We assume that the subjects'&nbsp;initial&nbsp;value is random.&nbsp;</p><p>In Model 2, We try to find a suitable error/covariance structure to explain the correlation&nbsp;among the values at each time points. So, besides selecting type of CS, you'd better try other type. such as UN, AR(1), <font face="Times New Roman">TOPE, HF</font>.. By comparing AIC or BIC, You can choose a final error/covaviance structuresult which fit the data most.</p><p></p>

<font face="Verdana" color="#000000">thanks zzheng722 and losttemple heip!</font>