<div dir="ltr"><div>



















<p class="MsoNormal" style="margin:0cm 0cm 0.0001pt"><span style="font-family:arial,helvetica,sans-serif"><font size="2">Hi all, <span></span></font></span></p><span style="font-family:arial,helvetica,sans-serif"><font size="2">

</font></span><p class="MsoNormal" style="margin:0cm 0cm 0.0001pt"><span style="font-family:arial,helvetica,sans-serif"><font size="2"><span> </span></font></span></p><span style="font-family:arial,helvetica,sans-serif"><font size="2">

</font></span><p class="MsoNormal" style="margin:0cm 0cm 0.0001pt"><span style="font-family:arial,helvetica,sans-serif"><font size="2">My question regards how to estimate overall magnitudes of
effect sizes from compiled studies regardless of the direction.<span>  </span>I have attached a figure to illustrate, which
I developed using made-up data and the attached code.<span>  </span><span></span></font></span></p><span style="font-family:arial,helvetica,sans-serif"><font size="2">

</font></span><p class="MsoNormal" style="margin:0cm 0cm 0.0001pt"><span style="font-family:arial,helvetica,sans-serif"><font size="2"><span> </span></font></span></p><span style="font-family:arial,helvetica,sans-serif"><font size="2">

</font></span><p class="MsoNormal" style="margin:0cm 0cm 0.0001pt"><span style="font-family:arial,helvetica,sans-serif"><font size="2">In the figure five studies have significantly positive
effect sizes, while 5 have significantly negative effect sizes.<span>  </span>Each have equal variances.<span>  </span>So, the overall estimated mean effect size
from a random effects model is 0.<span>  
</span>However, what if we simply want to estimate the mean effect size
regardless of direction (i.e. the average magnitude of effects)?<span>  In this example, that value would</span> be 9.58 (CI: 6.48, 12.67), correct?<span>  </span><span></span></font></span></p><span style="font-family:arial,helvetica,sans-serif"><font size="2">

</font></span><p class="MsoNormal" style="margin:0cm 0cm 0.0001pt"><span style="font-family:arial,helvetica,sans-serif"><font size="2"><span> </span></font></span></p><span style="font-family:arial,helvetica,sans-serif"><font size="2">

</font></span><p class="MsoNormal" style="margin:0cm 0cm 0.0001pt"><span style="font-family:arial,helvetica,sans-serif"><font size="2">I have heard that taking absolute values of effect sizes
generates an upward bias in estimates of the standardized mean difference.<span>  </span>Also, this would create a folded normal
distribution, which would violate assumptions of the model and would require an
alternative method of estimating confidence intervals. <span> </span>What would be your approach to setting up a
model for answering the question of how much the overall magnitude of responses
is?<span>  </span><span>  </span><span> </span><span></span></font></span></p><span style="font-family:arial,helvetica,sans-serif"><font size="2">

</font></span><p class="MsoNormal" style="margin:0cm 0cm 0.0001pt"><span style="font-family:arial,helvetica,sans-serif"><font size="2"><span> </span></font></span></p><span style="font-family:arial,helvetica,sans-serif"><font size="2">

</font></span><p class="MsoNormal" style="margin:0cm 0cm 0.0001pt"><span style="font-family:arial,helvetica,sans-serif"><font size="2">I suspect this question has come up in this email group in
the past.<span>  </span>If so, my apologies for the
redundancy, and please send me any reference that may be helpful.<span>  </span><span></span></font></span></p><span style="font-family:arial,helvetica,sans-serif"><font size="2">





</font></span><br clear="all"></div>Dave Daversa<br><div><br><div class="gmail_signature" data-smartmail="gmail_signature"><div dir="ltr"><div><div dir="ltr"><div><div dir="ltr"><div><div dir="ltr"><div><div dir="ltr"><div><div dir="ltr"><div><div dir="ltr"><div><div dir="ltr"><div><div dir="ltr"><div><div dir="ltr"><div><div dir="ltr"><div><div dir="ltr"><div><div dir="ltr"> </div></div></div></div></div></div></div></div></div></div></div></div></div></div></div></div></div></div></div></div></div></div></div></div></div></div>
</div></div>