microsimulation / ijm-xml

XML files for the International Journal of Microsimulation
MIT License
0 stars 0 forks source link

<boxed-text> in <p> in 00117 #183

Closed fred-atherden closed 4 years ago

fred-atherden commented 4 years ago

Article 00117 has a <boxed-text> element which is placed inside a <p> element.

<p>A key assumption is that there are no individual effects, that is, after taking into account other predictors, they are zero and all individuals have the same intercept. However, with longitudinal data the individual effects are generally not truly zero and so biased parameter estimates can occur. It is a well-documented phenomenon that the inclusion of an LDV in this manner violates the assumptions on the errors and results in biased estimates (<xref ref-type="bibr" rid="bib39">Verbeek, 2008</xref>:377–378; <xref ref-type="bibr" rid="bib9">Cameron and Trivedi, 2005</xref>:763–764).
<boxed-text id="box1">
<label>Box 1.</label>
<caption>
<title>Summary of OLS regression with a lagged dependent variable (LDV).</title>
</caption>
<table-wrap position="float">
<table frame="hsides" rules="groups">
<tbody>
<tr>
<td>Description of approach:</td>
<td>Standard model with an LDV.</td>
</tr>
<tr>
                    <td>Estimation:</td>
<td>Ordinary least squares (OLS)</td>
                  </tr>
<tr>
                    <td>Assumptions on individual effects and errors:</td>
<td>
                      <italic>μ<sub>i</sub>
                      </italic> = 0, <italic>ε<sub>i,t</sub>
                      </italic> = <italic>υ<sub>i,t</sub>
                      </italic>~<italic>iid N</italic>(0, <italic>σ</italic>
                      <sup>2</sup>), Cov(<italic>ε<sub>i,t</sub>
                      </italic>, <italic>
                        <bold>x</bold>
                        <sub>i,t</sub>
                      </italic>) = 0</td>
                  </tr>
<tr>
                    <td>Estimates of main effects of time-invariant predictors:</td>
<td>Estimated and included directly in the simulation</td>
                  </tr>
<tr>
                    <td>Implementation in the simulation:</td>
<td>Coefficients (<italic>γ</italic>, <italic>α</italic>, and <bold>
                        <italic>β</italic>
                      </bold>) applied to the corresponding individual values to get predicted mean for each individual. Random number drawn from normal distribution with mean equal to the predicted mean and standard deviation equal to <italic>σ</italic>.</td>
                  </tr>
<tr>
                    <td>Other advantages and disadvantages:</td>
<td>Simple to estimate the model and implement in the simulation.</td>
</tr>
</tbody>
</table>
</table-wrap>
</boxed-text>
</p>

Please ensure that this is instead placed as a child of the <sec>, i.e as below. Please also note that as per #159 this table should instead be captured as a <def-list>:

<p>A key assumption is that there are no individual effects, that is, after taking into account other predictors, they are zero and all individuals have the same intercept. However, with longitudinal data the individual effects are generally not truly zero and so biased parameter estimates can occur. It is a well-documented phenomenon that the inclusion of an LDV in this manner violates the assumptions on the errors and results in biased estimates (<xref ref-type="bibr" rid="bib39">Verbeek, 2008</xref>:377–378; <xref ref-type="bibr" rid="bib9">Cameron and Trivedi, 2005</xref>:763–764).</p>
  <boxed-text id="box1">
    <label>Box 1.</label>
    <caption>
      <title>Summary of OLS regression with a lagged dependent variable (LDV).</title>
    </caption>
    <table-wrap position="float">
      <table frame="hsides" rules="groups">
        <tbody>
          <tr>
            <td>Description of approach:</td>
            <td>Standard model with an LDV.</td>
          </tr>
          <tr>
            <td>Estimation:</td>
            <td>Ordinary least squares (OLS)</td>
          </tr>
          <tr>
            <td>Assumptions on individual effects and errors:</td>
            <td>
              <italic>μ<sub>i</sub>
              </italic> = 0, <italic>ε<sub>i,t</sub>
              </italic> = <italic>υ<sub>i,t</sub>
              </italic>~<italic>iid N</italic>(0, <italic>σ</italic>
              <sup>2</sup>), Cov(<italic>ε<sub>i,t</sub>
              </italic>, <italic>
                <bold>x</bold>
                <sub>i,t</sub>
              </italic>) = 0</td>
          </tr>
          <tr>
            <td>Estimates of main effects of time-invariant predictors:</td>
            <td>Estimated and included directly in the simulation</td>
          </tr>
          <tr>
            <td>Implementation in the simulation:</td>
            <td>Coefficients (<italic>γ</italic>, <italic>α</italic>, and <bold>
              <italic>β</italic>
            </bold>) applied to the corresponding individual values to get predicted mean for each individual. Random number drawn from normal distribution with mean equal to the predicted mean and standard deviation equal to <italic>σ</italic>.</td>
          </tr>
          <tr>
            <td>Other advantages and disadvantages:</td>
            <td>Simple to estimate the model and implement in the simulation.</td>
          </tr>
        </tbody>
      </table>
    </table-wrap>
  </boxed-text>
      </sec>