这个我还真是不熟悉。刚才下载了这个命令，查看了帮助文件，里面列出了一系列相关的参考文献，应该是可以顺藤摸瓜，找出背后的理论基础的。

Description

    dirifit fits by maximum likelihood a Dirichlet distribution to a set of variables
    depvarlist.  Each variable in depvarlist ranges between 0 and 1 and all variables
    in depvarlist must, for each observation, add up to 1: for example, they may be
    proportions.

    Note that cases will be ignored if the one or more of the dependent variables has
    a value less than or equal to zero or more than or equal to one or if the
    dependent variables don't add up to one.

    dirifit uses one of two parameterizations:

        A conventional parameterization with shape parameters alpha_j > 0 (one for
        each variable in depvarlist) (e.g. Evans et al. 2000 or Kotz et al. 2000) will
        be used if only depvarlist is specified or if one or more of alphavar() and
        alpha1|2|3|...|k() is specified.  alpha_j is reported on the logarithmic scale
        to ensure that it remains positive. The conventional parameterization is
        especially useful when no covariates are present.

        An alternative parameterization with location parameters mu_j (one for each
        variable in depvarlist except the baseoutcome) and scale parameter phi will be
        used if one or more of muvar(), mu1|2|3|...|k(), baseoutcome(), and phivar()
        is specified or if the alternative option is specified.  The alternative
        parameterization is especially useful when covariates are present. mu_j are
        reported on the multinomial logit scale so that they stay between 0 and 1, and
        add up to one. In order to help interpretation, various types of marginal
        effects can be calculated with ddirifit. phi is reported on the logarithmic
        scale to ensure that it remains positive. This parameterization is analogous
        to the parameterization proposed by Paolino (2001), Ferrari and Cribari-Neto
        (2004), and Smithson and Verkuilen (2006) for the beta distribution.

References

    Evans, M., Hastings, N. and Peacock, B. 2000. Statistical distributions.  New
    York: John Wiley.

    Ferrari, S.L.P. and Cribari-Neto, F. 2004.  Beta regression for modelling rates
    and proportions.  Journal of Applied Statistics 31(7): 799-815.

    Kotz, S., Balakrishnan, N., Johnson, N.L. 2000.  Continuous multivariate
    distributions: Volume 1. New York: John Wiley.

    MacKay, D.J.C. 2003.  Information theory, inference, and learning algorithms.
    Cambridge: Cambridge University Press (see pp.316-318).
    http://www.inference.phy.cam.ac.uk/itprnn/book.pdf

    Paolino, P. 2001.  Maximum likelihood estimation of models with beta-distributed
    dependent variables. Political Analysis 9(4): 325-346.
    http://polmeth.wustl.edu/polanalysis/vol/9/WV008-Paolino.pdf

    Smithson, M. and Verkuilen, J. 2006.  A better lemon squeezer? Maximum likelihood
    regression with beta-distributed dependent variables.  Psychological Methods
    11(1): 54-71.

跪谢连老师！！你真的是太好了，太伟大了，小猫感激不尽！！

其实我只找到几篇ｐａｐｅｒ，没有这么多资料，只是知道理论后面的基础是Dirichlet distribution。

再次感谢！