low`` end
      of the confidence interval is 0, and there is evidence against
      the null hypothesis at  the chosen `confidence_level` if
      ``high < OR``.
    * 'greater': the true odds ratio is greater than ``OR``.  The
      ``high`` end of the confidence interval is ``np.inf``, and there
      is evidence against the null hypothesis at the chosen
      `confidence_level` if ``low > OR``.

Returns
-------
ci : ``ConfidenceInterval`` instance
    The confidence interval, represented as an object with
    attributes ``low`` and ``high``.

Notes
-----
When `kind` is ``'conditional'``, the limits of the confidence
interval are the conditional "exact confidence limits" as described
by Fisher [1]_. The conditional odds ratio and confidence interval are
also discussed in Section 4.1.2 of the text by Sahai and Khurshid [2]_.

When `kind` is ``'sample'``, the confidence interval is computed
under the assumption that the logarithm of the odds ratio is normally
distributed with standard error given by::

    se = sqrt(1/a + 1/b + 1/c + 1/d)

where ``a``, ``b``, ``c`` and ``d`` are the elements of the
contingency table.  (See, for example, [2]_, section 3.1.3.2,
or [3]_, section 2.3.3).

References
----------
.. [1] R. A. Fisher (1935), The logic of inductive inference,
       Journal of the Royal Statistical Society, Vol. 98, No. 1,
       pp. 39-82.
.. [2] H. Sahai and A. Khurshid (1996), Statistics in Epidemiology:
       Methods, Techniques, and Applications, CRC Press LLC, Boca
       Raton, Florida.
.. [3] Alan Agresti, An Introduction to Categorical Data Analysis
       (second edition), Wiley, Hoboken, NJ, USA (2007).
)