<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Roger Day on Medium]]></title>
        <description><![CDATA[Stories by Roger Day on Medium]]></description>
        <link>https://medium.com/@rogerday_17101?source=rss-ae3762726ca1------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*fZtYOrrAVGTaGob5.</url>
            <title>Stories by Roger Day on Medium</title>
            <link>https://medium.com/@rogerday_17101?source=rss-ae3762726ca1------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Thu, 19 Nov 2026 17:36:20 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@rogerday_17101/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Dose Utility Explorer, a “shiny” app. Expected Utility, with Meihua and Marie and Devlish Merry]]></title>
            <link>https://medium.com/@rogerday_17101/dose-utility-explorer-a-shiny-app-expected-utility-with-meihua-and-marie-and-devlish-merry-da88cbe74af2?source=rss-ae3762726ca1------2</link>
            <guid isPermaLink="false">https://medium.com/p/da88cbe74af2</guid>
            <category><![CDATA[r]]></category>
            <category><![CDATA[pharmacology]]></category>
            <category><![CDATA[expected-utility]]></category>
            <category><![CDATA[medicine]]></category>
            <category><![CDATA[shiny-app]]></category>
            <dc:creator><![CDATA[Roger Day]]></dc:creator>
            <pubDate>Thu, 03 Apr 2025 17:45:47 GMT</pubDate>
            <atom:updated>2025-04-03T17:45:47.032Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*1bgyaP9K_enmWwTg8mzmzQ.png" /></figure><p><a href="https://www.youtube.com/watch?v=c2IXn-rBl6k">https://www.youtube.com/watch?v=c2IXn-rBl6k</a></p><p>There’s a new medicine. It might help my problem. Problem is, no one has ever taken this medicine before. It could help but it could harm. We’ll touch on individual feelings of hope and fear. The hope that the medicine will give you a good response, and the fear that it will give you some bad toxic outcome. Because these constitute personal feelings and values.</p><p>Help or harm could depend a lot on the dose.</p><p>The <strong>Dose Utility Explorer</strong> is an app that Meihua Wong and Marie Gerges made with me awhile back. You can play with it yourself; it’s it’s live at <a href="https://trials.shinyapps.io/DUEshiny/">DUEshiny/ </a>. The <a href="https://www.youtube.com/watch?v=DskXaA-nqoI">corresponding video</a> walks you through, not only the concepts, but also the app itself.</p><p>The key ideas are</p><ul><li>average utility,</li><li>dose thresholds based on pharmacokinetics and pharmacodynamics, and</li><li>what do you do when you’re missing a lot of important information?</li></ul><p>We uncovered some crazy things about dose utility curves and Phase One clinical trials. Remember when people said Phase One trials try to find a dose where a third of people experience dose-limiting toxicity? Hmm, how good a rule of thumb is that?</p><p>Based on a decision support framework called Bayesian statistics. Reverend Thomas Bayes. Non conformist reverend from the 1700s who is still revolutionizing computer science, statistics, data analysis, and much more.</p><p>Featuring also some music from the unique Pittsburgh folk band <a href="https://www.youtube.com/watch?v=DskXaA-nqoI">Devlish Merry</a>.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=da88cbe74af2" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Harris campaign’s political consultants and missed opportunities]]></title>
            <link>https://medium.com/@rogerday_17101/harris-campaigns-political-consultants-and-missed-opportunities-688e647cdaa4?source=rss-ae3762726ca1------2</link>
            <guid isPermaLink="false">https://medium.com/p/688e647cdaa4</guid>
            <category><![CDATA[election2024]]></category>
            <category><![CDATA[history]]></category>
            <category><![CDATA[political-consultants]]></category>
            <dc:creator><![CDATA[Roger Day]]></dc:creator>
            <pubDate>Sat, 16 Nov 2024 02:39:19 GMT</pubDate>
            <atom:updated>2024-11-16T02:39:19.023Z</atom:updated>
            <content:encoded><![CDATA[<p>The Harris campaign missed valuable opportunities because their political consultants, every last one of them, are timid, unimaginative and ignorant of things outside the media.</p><p>Here are five .</p><p>Missed opportunity #1</p><p>Peace activists Y<a href="https://www.democracynow.org/2024/10/7/the_path_forward">AEL<strong> </strong>BRAUDO<strong>-</strong>BAHAT and MAOZ<strong> </strong>INON</a>, one Palestinian, one Israeli, lost close family to the brutality of this war.</p><p>Imagine if the Democratic Convention had allowed these two to speak jointly, with Harris saying, I may not agree with everything these courageous people have to say, but their voices and voices of peace need to be heard much more.So I am giving this opportunity.</p><p>She would be admired for her courage. Instead, ironically large numbers of both American Palestinians and American Jews were equally repelled.</p><p><strong>Missed opportunity #2</strong></p><p>Fracking.</p><p>Harris switched to a fracking supporter. Why? Never clear. Ironically large numbers of both supporters and opponents of fracking have been equally repelled.</p><p>She could have said, fracking technology will be an essential element in the development of geothermal energy. Always on. No need for constant fuel supply, so always free once the infrastrucure is built. With fracking, available in great areas across the nation.</p><p><strong>Missed opportunity #3</strong></p><p>“We’re not going back.”</p><p>A slogan devised as a reflex answer “make America great again”, because innovation is not what political consultants do. Muddy. Anyone can interpret that in a hundred different ways. What did Harris mean by it?</p><p>A lost opportunity. Because, YES we do want to go back in important ways, important to state. Here are a few.</p><p>* 1980 when economic inequality was a lot smaller. <a href="https://www.motherjones.com/politics/2024/11/why-donald-trump-won-election-white-house/">https://www.motherjones.com/politics/2024/11/why-donald-trump-won-election-white-house/</a></p><p>* When McCain-Feingold, and many other campaign finance limits had not yet been decimated.</p><p>* When the Voting Rights Act was in full force before SCOTUS crippled it.</p><p>* When the climate was not so dangerous, and the natural environment was in better shape: oceans, soils, waters, air, species.</p><p>* When the individual, not the legislators and governers, decided about abortion.</p><p>* When high powered weapons were banned.</p><p><strong>Missed opportunity #4</strong></p><p>“Danger to democracy” “Fascist”</p><p>Just words. You’re never going to win a name-calling battle with Trump.</p><p>T<strong><em>each history</em></strong>. Most people don’t know much. Teach good examples of patriotism that promote the beauty of peaceful transition .</p><p>Adams vs Jefferson 1800, a bitter election, a peaceful transition.</p><p>Al Gore confirming his opponent George Bush despite getting far more votes.</p><p>Hillary conceding immediately.<br>Joe Biden confirming Trump, despite Hillary getting far more votes.</p><p>Obama giving the Trumps a tour of the White House.</p><p>Where was the ad showing Gore confirming Bush? <br>One of the noblest moments in American history.</p><p><strong>Missed opportunity #5</strong></p><p>Celebrities.</p><p>Who gives a shit about celebrities?</p><p>Instead, give oxygen to Rev William Barber and others who fight for working people, struggling people, homeless people, people forced to breathe the air of industrial parks spewing poisons.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=688e647cdaa4" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Wilcoxon two-sample test: ties make a mess of things…what to do?]]></title>
            <link>https://medium.com/@rogerday_17101/wilcoxon-two-sample-test-ties-make-a-mess-of-things-what-to-do-ab9c2c7e36e6?source=rss-ae3762726ca1------2</link>
            <guid isPermaLink="false">https://medium.com/p/ab9c2c7e36e6</guid>
            <category><![CDATA[wilcoxon]]></category>
            <category><![CDATA[multiple-testing]]></category>
            <category><![CDATA[ties]]></category>
            <category><![CDATA[statistics-analysis]]></category>
            <dc:creator><![CDATA[Roger Day]]></dc:creator>
            <pubDate>Sun, 18 Aug 2024 15:07:00 GMT</pubDate>
            <atom:updated>2024-08-18T15:07:38.252Z</atom:updated>
            <content:encoded><![CDATA[<p>Trigger warning: this article uncritically presumes the use of “P values” in comparing two groups of numbers.</p><p>There are loads of reasons to despise P values; elsewhere I have a polemic chanting a few of those reasons.<br>It also presumes that multiple testing adjustments are in play. That’s got some ridiculous paradoxes too. And a corresponding polemic. All in good time on the <a href="https://www.youtube.com/@statisticsbeautiful1098">StatisticsBeautiful</a> channel.</p><p>So… let’s just agree to get along for the next few minutes.</p><p>An old old work horse of “classical” frequentist statistical analysis, the Wilcoxon test comes in two forms. This will be about the two-sample form, comparing two collections of numbers, to see if one tends to be bigger than the other (essentially, are the medians different?). Those numbers constitute a “feature”. The Wilcoxon test gets employed to judge whether the feature tends to be higher in one sample, one group, than the other.</p><p>What is a “feature”? A number. Statisticians used to call them “covariates”. Not classy. Better PR in computer science.</p><p>The testing tool is the function wilcox(), a basic workhorse in the statistical language R. Why was it necessary to cheat poor Mr. Wilcoxon out of the last two letters of his name? DM me if you know!</p><h4>The problem with ties</h4><p>The basis for the test is to count the number of pairs, one from group #1 and the other from group #2, for which the group #2 number is larger.</p><p>When there are many many features to evaluate (such as high-throughput genomic data), P values come spitting out in great numbers, and many very small P values can be generated just by chance. So one becomes skeptical of “discoveries”, and requires extra strong evidence, an exceptionally tiny P value. That P value can be computed with a approximations, or exactly based on the randomization distribution. The exact calculation has more prestige. (Though, like many frequentist calculations, the sampling distribution assumed is not nature’s sampling distribution, so “exact” is not exactly true.)</p><p>But when the data contain ties, the exact distribution of the statistic is much more difficult and the approximations are not good. This doesn’t matter much for P values &gt; 0.001, but can be a huge effect for very very small P values. Very very small P values matter <em>a lot</em> when the number of features, the number of comparisons, is huge.</p><h3>Numerical experiment</h3><p>In this numerical experiment, I create a data set with 100 observations and 300 features, and one binary label per observation; no imagination here, so the labels are #1 and #2.</p><p>For each feature, the effect size is chosen randomly, from a normal distribution with a standard deviation of 2. That effect is added to the group #2 observations.</p><p>Let’s look at three ways that wilcox can calculate the P value: normal approximation (default), correct=TRUE, and exact=TRUE.</p><p>(Want to see the R markdown which created this document? Look for the file wilcoxon-ties-experiment.Rmd at <a href="https://github.com/professorbeautiful/bioinf2118">https://github.com/professorbeautiful/bioinf2118</a>. In R markdown, the warning messages have been suppressed using the chunk option <strong>{r warning=FALSE}</strong>. Warnings are so bourgeois.)</p><p>Now we combine the results across all the 300 features. Here the rounding takes the 300 features and creates ties in 199 of them, with 101 still without ties.</p><h4>Figure 1</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/864/1*-5g9b7XMptXs4f2hq_4jgA.png" /></figure><p>In Figure 1, each point is a feature, with the effect size chosen randomly. The “green T” is a feature that has at least one tie. The “red O” is a feature that has no ties. The graph compares to approximation values to the correct values. Excuse me, the correct=TRUE P values. “Correct” does not mean they are correct. It means only that a “continuity correction” for the normal approximation was applied, to adjust (roughly) for the discreteness of the Wilcoxon test statistic. Sounds good; but makes no discernable difference, whether there are ties in the feature or not.</p><h3>The Exact-o knife</h3><p>Let’s instead compare the approximation to the exact P value. Excuse me, the exact=TRUE P value. Surprise, it may not be exact, as we’ll see.</p><h4>Figure 2</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*hfroNVJAAaUWwdaP-vdkrA.png" /></figure><p>Now this is strange! For features with ties, the green T’s agree perfectly like before. But for those with no ties, when the exact P value is really small, the approximation is terrible, hugely bigger, less significant, than it should be.</p><p>This matters <em>a lot</em>, because when there are many many comparisons (high-throughput biological data), small P values are frequent, including small P values generated just by chance. People apply very stringent P value cutoffs to reduce the risk of false positive “discoveries” that are just artifacts of chance.</p><p>What’s going on??</p><h3>The effect of ties</h3><p>When the data contain ties, the exact distribution of the statistic is much MUCH more computationally intensive, and wilcox.test simply will not do it. Instead it just puts out the approximation. (There is a warning message, but as noted above we have suppressed it.)</p><p>Let’s look at this by effect size. For the approximate P values, ties and no-ties features follow the same trajectory.</p><h4>Figure 3</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*w1oZL7FwR1NBue5GZdPRdg.png" /></figure><p>But for the “exact” computation, the presence of ties prevents the “exact” method from being used.</p><h4>Figure 4</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*CYSIQRsZdywq9Kmnk9racA.png" /></figure><h4>Studying one feature.</h4><p>To illustrate, we create a single feature with a pretty strong effect size = 2. We do a no-tie version, but where the second value is replaced by X[2] = X[1] + 0.0001, very close to a tie. Then we successively round the data to fewer and fewer digits.</p><p>Changing the rounding</p><pre>digits     p.approx    p.correct      p.exact anyTied<br>      6 6.528263e-12 6.528263e-12 5.321034e-14   FALSE<br>      5 6.528263e-12 6.528263e-12 5.321034e-14   FALSE<br>      4 6.528263e-12 6.528263e-12 5.321034e-14   FALSE<br>      3 6.371475e-12 6.371475e-12 6.371475e-12    TRUE<br>      2 6.520726e-12 6.520726e-12 6.520726e-12    TRUE<br>      1 5.802984e-12 5.802984e-12 5.802984e-12    TRUE<br>      0 6.420794e-12 6.420794e-12 6.420794e-12    TRUE</pre><p>Look at the “p.exact” column. Once you round to 3 or fewer digits, ties appear. And all hell breaks loose. In the sense that the “exact” calculation is no longer done, but the approximation substitutes for it.</p><p>Is that good? Well, if you don’t mind increasing the P value by orders of magnitude… But if you are committed to coping with multiple testing with some “significance” adjustment like the Bonferroni method, then that is a disaster.</p><p>Elsewhere, I have a long rap about the perils of the P value, and another long rap about the perils of multiple testing, so … calm down.</p><h3>Breaking the ties</h3><p>Now we will take the digits=3 rounded feature, but break the ties, by adding a tiny increment to a duplicated value. That should give a more appropriately small P value.</p><pre>Warning in wilcox.test.default(X1, X2, exact = T): cannot compute exact p-value<br>with ties</pre><p>Oops, I forgot to suppress the warning message this time. But… Aha! with the original data with ties, the P value is 6.4^{-12}, but with the teensy breaking of the ties, it’s 5.3^{-14}. Now the exact P value is orders of magnitude smaller, as it should be. So we have regained the power lost when the ability to do the exact permutation was lost.</p><p>The documentation for wilcox says:</p><pre>For stability reasons, it may be advisable <br>to use rounded data or to set digits.rank = 7, say, ...</pre><p>That might not be such good advice.</p><h3>Computing time</h3><p>The computing time increases somewhat due to <em>exact=TRUE</em>, when there are no ties. <em>If there are ties, the option is ignored</em>. We suppress the warnings again.</p><p>For looping over 200 repetitions, the computation takes 319 milliseconds for the “rounded to 3 digits” feature, with its ties, and 534 milliseconds when the ties are broken by tiny arbitrary amounts.</p><p>There is an alternative to the built-in wilcox.test function. In the R package <a href="https://search.r-project.org/CRAN/refmans/coin/html/00Index.html">coin</a>, the function wilcox_test is worth a look (notice the underscore instead of the dot; cool, … but still the &quot;on&quot; is missing!? No matter; cyberspace shows no WilcoxOFF).</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ab9c2c7e36e6" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Slam Poet Statisticians]]></title>
            <link>https://medium.com/@rogerday_17101/slam-poet-statisticians-1daeb562f533?source=rss-ae3762726ca1------2</link>
            <guid isPermaLink="false">https://medium.com/p/1daeb562f533</guid>
            <category><![CDATA[poetry]]></category>
            <category><![CDATA[bayesian-statistics]]></category>
            <category><![CDATA[r]]></category>
            <category><![CDATA[prior]]></category>
            <category><![CDATA[limerick]]></category>
            <dc:creator><![CDATA[Roger Day]]></dc:creator>
            <pubDate>Sun, 18 Aug 2024 01:07:57 GMT</pubDate>
            <atom:updated>2025-04-03T17:40:52.494Z</atom:updated>
            <content:encoded><![CDATA[<blockquote>There was an old frequentist <strong>bloke</strong></blockquote><blockquote>Who thought that old Bayes was a <strong>joke</strong></blockquote><blockquote>“Subjectivity <strong>stinks</strong>!</blockquote><blockquote>The prior’s a <strong>jinx</strong>!</blockquote><blockquote>Your posterior’s mirrors and <strong>smoke</strong>!”</blockquote><blockquote>Tom Bayes said “My rear end is <strong>fine</strong>,</blockquote><blockquote>The chances are eight out of <strong>nine</strong>.</blockquote><blockquote>Quit the <strong>invective</strong>,</blockquote><blockquote>You’re just as <strong>subjective</strong>.</blockquote><blockquote>But <em>your</em> prior’s hidden. Not <strong>mine</strong>.</blockquote><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=1daeb562f533" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Descript and Camtasia- how to enjoy the best of both worlds]]></title>
            <link>https://medium.com/@rogerday_17101/descript-and-camtasia-how-to-enjoy-the-best-of-both-worlds-12a0c9445996?source=rss-ae3762726ca1------2</link>
            <guid isPermaLink="false">https://medium.com/p/12a0c9445996</guid>
            <category><![CDATA[r]]></category>
            <category><![CDATA[camtasia]]></category>
            <category><![CDATA[editing]]></category>
            <category><![CDATA[vi̇deo]]></category>
            <category><![CDATA[descript]]></category>
            <dc:creator><![CDATA[Roger Day]]></dc:creator>
            <pubDate>Fri, 10 May 2024 15:21:52 GMT</pubDate>
            <atom:updated>2025-12-02T17:24:50.236Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/358/1*KZXMqNFgOlXSg_ABzRGjrw.jpeg" /></figure><p>Camtasia is great for making videos with fancy video elements, like photos, animations, transitions, …. I use it to make my <a href="http://www.youtube.com/professorbeautiful">music videos </a>even sillier.<br>But Camtasia is <em>terrible terrible</em> at spoken audio that you want to edit. (The Audiate addon is expensive and weak).</p><p>Descript is marvelous at spoken audio , with many cool features. [Ed: I’m even using “underlord”! When I”m too lazy to find the right menu pick.]</p><p>And organization! Search for text in transcript and marker labels, <br>see all the hits at once, scroll a list of all the markers and <br>go where you want to go…<br>(I used to like the concept of “scenes”, but soured on it.)</p><p>But Descript is <em>terrible terrible</em> at including visual features,<br>The capabilities are profoundly limited and awkward to use.</p><p>(I used Adobe Premiere for a while, extremely powerful, but couldn’t pump enough glucose into my few remaining brain cells to cope with the arcane interface.)</p><p>So how do we use the best of both D and C? (Please excuse the abbreviations.)</p><p>Solution in “tldr” form:</p><p>1. Create and edit the script and voice in Descript. [Edit: desktop just busted bad. So I use the web page version. It’s been exceptionally good.]</p><p>2. Insert Descript markers — lots of them — to label sections and make notes for desired visual elements.</p><p>3. “Publish” and export to an audio (or video) file.</p><p>4. Pull that into a Camtasia project and</p><p>5. This is the key step: copy the markers to your Camtasia project, automatically. Then in C add lots of wonderful video elements, guided by the markers.</p><p>The details for the last step were pretty arcane to work out.<br>But it works! And now, it’s yours. Well, there may be a little sweat first.<br>The payoff is big:</p><p># Navigate in D via search (which you can’t do in C). Then easily move to the corresponding time in C. Create the visual content at that time.</p><p># Select a section in D, you’ll see immediately how long is its duration (which you can’t do in C).</p><p># To play around with visual elements in C, check the time in C and easily move to the corresponding time in D to read the neighboring transcript text &amp; figure out what to do with the visual elements. [Edit: You can also import the srt caption track into C…. but the caption feature is severely stupid in C.]</p><p># Make an alteration in D to the narration or the markers, then transfer it to C without messing things up. (This is quick! Except for waiting for the publish-update-and-download to complete. Edit: actually, using D on the web, it’s damn fast.)</p><p>Let’s dive into the details for each step. (A couple other kludges for coordinating D and C timelines will be appended.)</p><p>1. You can <strong>write a script and record it</strong> into D, or let D translate it into a stock AI voice they provide or an AI voice that you have “trained” that sounds like you. For me, that voice is lacking in liveliness, so I end up replacing it with spot recordings of my own voice. A delight here is the ability to edit the text and audio, and especially to adjust silent spaces between words. D shows how big a gap or a selected section is, in milliseconds, so easy to adjust.</p><p>2. <strong>Make a marker in the script</strong> whereever you will want a marker to appear in your Camtasia project. (A Descript marker will become a Camtasia marker, as you’ll see.) You type “#” and enter a title. (Only one line per marker.) It’s handy to add a note describing what you will want to add at that spot in the final video, such as an image or annotation group. [I edited this paragraph: I used to recommend also creating “<strong>scenes”</strong> by typing “/” , but that doesn’t seem to contrbute much.]</p><p>3. “<strong>Publish</strong>” will upload to the D cloud. With “update” a new version will be created in the background by D’s computers, not yours. Save the link, and colleagues can get the new version (as soon as it’s done processing).</p><p><strong>Download</strong> the video to the local drive; the folder where it is saved, call it <strong>projectLocation</strong>; later it will be used in a variable.</p><p>4. <strong>Create a Camtasia project</strong> in <strong>projectLocation</strong>. Its name will will be used in a variable called <strong>cmProjName</strong>. Using the Media bin, or File/Import media, pull in the video.</p><p>5. Here’s the big enchilada: <strong>transporting the markers</strong>. Hang on. This is good.</p><p>In Descript, go to Publish. Switch to the Export tab.</p><p>Export Transcript with Format = MARKDOWN, toggle on “Include Markers” . Also, under Timecodes toggle “Markers” on.</p><p>Make sure that your Camtasia project is <strong>closed</strong>.</p><p>If you have the program “R” installed, open the script “copy_markers_from_D_to_C.R”, edit it to set the variables “projectLocation” and “cmProjName”, and run it. That script is at <a href="https://github.com/professorbeautiful/DUE/blob/master/R/copy_markers_from_D_to_C.R">https://github.com/professorbeautiful/DUE/blob/master/R/copy_markers_from_D_to_C.R</a>. [Ed.: CORRECTED THE LINK] Then run (“source”) the file copy_markers_from_D_to_C.R, and run the function copy_markers_from_D_to_C().</p><p>(If you don’t want to bother with “R”, but you’re good at python or cobol or whatever, it should be fairly easy to read and follow and rewrite this R code in the language of your choice. A bit complicated because the representation of marker times requires conversion from D to C formats.)</p><p>This extracts the marker info from the D transcript file, reads the Camtasia tscproj file “project.tscproj”, copies it to a backup with “BACKUP” and a time stamp inside the cmproj folder. It then inserts the marker info in the proper place to create a string “newFile”. <strong>Markers previously on the Camtasia timeline, if any, are totally replaced by the new markers from D. Markers attached to media or other elements on tracks in C are retained.</strong></p><p>Then it writes newFile over “project.tscproj”.</p><p>Now you can reopen the Camtasia project. Voilà. (Until Camtasia changes its tscproj file format!)</p><p>=-=-=-=-=-=-=-=-=-=-=</p><p><strong>Now what? Coordinating the editing process</strong></p><p>Now you can do the video editing in Camtasia using the powerful tools there; quicker and far more flexible. adding elements (like annotations, pictures, screen captures etc.), whatever makes you happy. (I must say, though, that C does not allow you to search the Media bin for a media import by name; D does.)</p><p>BUT… navigating to a marker is a pain, in both programs. In Descript, you can search text and that includes markers, though not the markers only. Or scroll through a not-searchable dropdown of markers. That’s acceptable. Worse is Camtasia. There is no list of markers and no searching them.</p><p>This you can do: find your marker in Descript, note the time on the timeline, specifically the minutes and seconds. Then in Camtasia, look for the time box, to the right of the scroll bar under the canvas. Click to open the time box, click again in the box to make it receptive to input, and type in the minutes, seconds, and two zeros, with no separators. (Despite appearances, if you put in colons or semicolons or whatever, it will not work.) Try not to worry about fractions of a second. “Don’t sweat the small stuff.”</p><p>Conversely, if you are in Camtasia and want to make your way to the same spot in Descript, look at the time box. Note the minutes and seconds. In Descript, click Actions, then Jump to…, then Jump to Time… What, no shortcut?? Yup, no shortcut. I use Keyboard Maestro to create my own.</p><p>Enter minutes, then colon (:), then seconds. In D, you DO need the separator. Again, try not to sweat the fraction of a second. It’s possible (frames in C vs milliseconds in D) but not worth the trouble.</p><p>(You could export only the video from Camtasia, then import into Descript, not clear when or if that’s going to work well. Not my choice.)</p><p>Everyone says, finalize the script before working on the video part, and hope you don’t have to edit the script again, which sends you into synchronization hell.<strong> </strong>But wait, what if I want to go back and make changes in the Descript project, inserting, replacing, … and some audio gets pushed around, lengthened or shortened? That’s ok. Just follow the same *simple* steps, but with a couple of extra steps inserted, as indicated here in italic:</p><p>In Descript re-publish, and download the video, and re-export the transcript. <br><em>In Camtasia project, move markers from the main timeline to the individual elements (like annotations, pictures, screen captures etc..) on their tracks.</em> <br>Close the Camtasia project. <br>Re-run that R function copy_markers_from_D_to_C() (or your translation of it). <br>Re-open the Camtasia project. <br><em>Delete the previous video track</em>, and import the new video (CMD-I, or use the Media bin). <br>Place and align the video . Align the elements with their legacy trackmarkers to the newly imported markers. <br>You’re all set.</p><p>Then take a nap.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=12a0c9445996" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Gentle (but firm) Intro to Statistics via Medical Decision-making for Mammography]]></title>
            <link>https://medium.com/@rogerday_17101/gentle-but-firm-intro-to-statistics-via-medical-decision-making-for-mammography-77027a0b20db?source=rss-ae3762726ca1------2</link>
            <guid isPermaLink="false">https://medium.com/p/77027a0b20db</guid>
            <category><![CDATA[mammography]]></category>
            <category><![CDATA[bayesian-statistics]]></category>
            <category><![CDATA[positive-predictive-value]]></category>
            <category><![CDATA[videos]]></category>
            <category><![CDATA[shiny]]></category>
            <dc:creator><![CDATA[Roger Day]]></dc:creator>
            <pubDate>Fri, 19 Apr 2024 02:32:43 GMT</pubDate>
            <atom:updated>2024-07-03T13:50:06.221Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/978/1*GhYaUj3_-KkHSSh--V41vw.png" /></figure><p>You may be shocked to learn that the average gynecological oncologist has not a clue how to interpret what a mammography result says about risk. Most are wildly off. Check out this great article by Gerd Gigerenzer et al, “<a href="https://journals.sagepub.com/doi/10.1111/j.1539-6053.2008.00033.x">Helping Doctors and Patients Make Sense of Health Statistics”</a>. The perspective laid out there inspired what follows. (It also contributed to an article of mine on biomarker studies. That’ll wait for another day.)</p><p>What are the factors affecting the decision to treat or not treat a patient with a positive medical test result, like a positive mammography test?</p><p>You can learn about them, and how to think clearly about decisions as important as what to do with a positive mammogram, with my interactive medical decision app “<a href="https://trials.shinyapps.io/simplebayesdecision-fancy/">Simple Bayes decision plots</a>”. <br>It’s pretty easy to use. No, really,<br>And there are four different displays to suit your taste.</p><p>A gentle video tour, plus a little exploration of the ideas: check out<br><a href="https://www.youtube.com/watch?v=34IH5hz5ul4">Bayes medical decision-making: explore the app: Statistics Beautiful Episode 2</a> (with tuba/bohdran duo musical intro).<br>Or, if you prefer a no-youtube version, <a href="https://professorbeautiful.org/videos/video-gallery-code/gallery/Simple_Bayes_Treatment_Decision.mp4">version-with-no-social-media-frills</a>.</p><p>In the video, the app demonstrates how to think about this clinical mammography problem, and reviews the math a little. (Addition, multiplication, division; that’s it. ) The key idea is “<strong>Bayes expected loss</strong>”, which thrives as a driving principle in decision theory in general. The story of <a href="https://en.wikipedia.org/wiki/Thomas_Bayes">Reverend Thomas Bayes</a>, officially a “Non-conformist”, goes back to the late 1700s. Almost forgotten for a century and a half, but now at the core of many many scientific and engineering fields. (But more on that another day.)</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=77027a0b20db" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>