<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:dc="http://purl.org/dc/elements/1.1/">
  <channel>
    <title>DEV Community: Iqbal Taher</title>
    <description>The latest articles on DEV Community by Iqbal Taher (@iqbaltaher).</description>
    <link>https://dev.to/iqbaltaher</link>
    <image>
      <url>https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F4210742%2F5c216c48-3b55-4a82-a380-c634f3a5600f.jpg</url>
      <title>DEV Community: Iqbal Taher</title>
      <link>https://dev.to/iqbaltaher</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/iqbaltaher"/>
    <language>en</language>
    <item>
      <title>What 625,987 NHTSA complaints look like once you count them one way</title>
      <dc:creator>Iqbal Taher</dc:creator>
      <pubDate>Tue, 27 Oct 2026 11:49:07 +0000</pubDate>
      <link>https://dev.to/iqbaltaher/what-625987-nhtsa-complaints-look-like-once-you-count-them-one-way-f72</link>
      <guid>https://dev.to/iqbaltaher/what-625987-nhtsa-complaints-look-like-once-you-count-them-one-way-f72</guid>
      <description>&lt;p&gt;Every car owner in the US can file a safety complaint with NHTSA, and every one of those complaints is public. The catch is the format. The flat file is about 1.6 GB of tab-separated text, one row per complaint &lt;em&gt;per component&lt;/em&gt;, written for investigators rather than for anyone trying to answer a simple question like "which models draw the most complaints?"&lt;/p&gt;

&lt;p&gt;I wanted that answer for model years 2010 to 2025, and I wanted it in a form other people could check and cite. Here is what I learned turning the raw file into a small, citable dataset.&lt;/p&gt;

&lt;h2&gt;
  
  
  One complaint, one vehicle, counted once
&lt;/h2&gt;

&lt;p&gt;The first trap is double counting. A single complaint can sit on several rows, because the owner ticked more than one component (say, "steering" and "electrical system"). If you count rows, a busy complaint looks like three.&lt;/p&gt;

&lt;p&gt;So the rule is simple and written down before any counting starts:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Each complaint is counted &lt;strong&gt;once per make, model and model year&lt;/strong&gt;, keyed on its complaint number.&lt;/li&gt;
&lt;li&gt;It is counted &lt;strong&gt;once per component&lt;/strong&gt; it names, so component totals can add up to more than the vehicle total.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Writing the rule first matters more than the code. It is what lets someone else rebuild the numbers and get the same result.&lt;/p&gt;

&lt;h2&gt;
  
  
  Keep NHTSA's own names
&lt;/h2&gt;

&lt;p&gt;It is tempting to tidy model names that appear under several spellings, or merge sub-components. I didn't. Model names and components stay exactly as NHTSA records them. Merging feels cleaner, but every merge is a judgment call someone else can't reproduce. When I tested merging the common name variants anyway, none of the top 25 rows moved.&lt;/p&gt;

&lt;h2&gt;
  
  
  Counts are not failure rates
&lt;/h2&gt;

&lt;p&gt;The file has no sales figures. A model that sold in huge numbers can collect more complaints than a rarer car with a worse record. That is the single most important caveat, so it goes in the dataset card, the report and the CSV notes, not in a footnote.&lt;/p&gt;

&lt;h2&gt;
  
  
  What came out
&lt;/h2&gt;

&lt;p&gt;From the copy of the file dated August 25, 2026:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;625,987 complaints across 4,087 make, model and model-year combinations.&lt;/li&gt;
&lt;li&gt;The 2010 Ford Fusion has the most of any vehicle, 5,126, with steering named most.&lt;/li&gt;
&lt;li&gt;Ford holds 17 of the 25 most complained-about vehicles.&lt;/li&gt;
&lt;li&gt;The only vehicle from model year 2023 or newer in the top 100 is the 2023 Tesla Model Y.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Making it citable
&lt;/h2&gt;

&lt;p&gt;A dataset nobody can cite is a dataset nobody uses. So the CSV (one row per make, model, model year and component, 90,254 rows), a JSON summary and the build script are archived on Zenodo with a DOI (10.5281/zenodo.23685716), and mirrored on GitHub, Hugging Face and Kaggle under CC BY 4.0. Each edition stays frozen at its own address, so a number quoted today still points to the same number next year.&lt;/p&gt;

&lt;p&gt;If you want the charts, the full ranking and the method written out for non-developers, the write-up is here: &lt;a href="https://oilandmiles.com/reports/most-complained-cars" rel="noopener noreferrer"&gt;the most complained-about cars ranked by NHTSA complaints&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;&lt;em&gt;I'm Iqbal Taher. I write for &lt;a href="https://oilandmiles.com" rel="noopener noreferrer"&gt;Oil &amp;amp; Miles&lt;/a&gt;, where we turn federal car records into plain-English guides.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>opendata</category>
      <category>datascience</category>
      <category>python</category>
      <category>dataset</category>
    </item>
  </channel>
</rss>
