--- title: Building Models in Two Worlds: From Latent Constructs to... description: A data scientist compares modeling latent constructs (survey-based SEM in academic research) versus behavioral signals (ML-based propensity and lift models in... canonical: https://daily.dev/posts/building-models-in-two-worlds-from-latent-constructs-to-behavioral-signals-uroobl2k7 twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Building Models in Two Worlds: From Latent Constructs to Behavioral Signals | daily.dev og:description: A data scientist compares modeling latent constructs (survey-based SEM in academic research) versus behavioral signals (ML-based propensity and lift models in... og:url: https://daily.dev/posts/building-models-in-two-worlds-from-latent-constructs-to-behavioral-signals-uroobl2k7 og:image: https://api.daily.dev/og/posts/uroobl2K7.png og:image:alt: Building Models in Two Worlds: From Latent Constructs to Behavioral Signals og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Building Models in Two Worlds: From Latent Constructs to Behavioral Signals **[Towards Data Science](https://daily.dev/sources/tds)** · 18 min read · 0 upvotes · 0 comments ## Summary A data scientist compares modeling latent constructs (survey-based SEM in academic research) versus behavioral signals (ML-based propensity and lift models in industry ad measurement). The core insight is that the statistical machinery is nearly identical, but the context inverts almost every decision: where variables come from, what makes a variable valid, how features are selected, and what the model is expected to explain. Key practical lessons include treating nonsensical estimates as diagnostics rather than results, maintaining a proxy-to-construct mapping for behavioral features, understanding when multicollinearity matters (interpretation vs. prediction), and validating pre-period alignment against seasonal patterns before trusting lift estimates. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### Why did a geo lift test show that an advertising campaign caused negative sales for a snack-category product? A seasonality mismatch between test and comparison regions caused the false negative reading, not an actual suppression effect from the ads. The product's sales ramp toward a seasonal peak at different times across regions, so during the campaign window the test region was still climbing while comparison regions had already crested, and after the campaign the test region read below comparison, mimicking an ad-caused decline. The real failure was the assumption that groups would have moved together absent the ad. _Anyone debugging a suspicious lift or causal-inference result can find similar diagnostic breakdowns curated on daily.dev._ ### Why is correlation among features good in a structural equation model but bad in a propensity or predictive model? In structural equation modeling, correlated survey items within a construct are the evidence that they measure one underlying latent variable, so high internal correlation signals good reliability. In predictive models like propensity classifiers, that same redundancy among features destabilizes linear coefficients and smears feature importance across correlated tree splits, so it's treated as a problem rather than a virtue when interpretation matters. _Developers weighing multicollinearity trade-offs across modeling approaches can track similar analysis on daily.dev._ ### How do you validate a latent construct like privacy concern or perceived usefulness before using it in a regression model? Build the construct from several survey indicators using structural equation modeling rather than averaging items directly, then check that items load properly on their intended construct, that the construct holds together internally via Cronbach's alpha or composite reliability, and that it explains more variance in its own indicators than it shares with other constructs via average variance extracted and discriminant validity. Averaging blindly lets measurement error bias coefficients toward zero. _Practitioners building measurement models can find related methodology discussions on daily.dev._ ## Similar posts on daily.dev - [From Prediction to Intervention](https://daily.dev/posts/from-prediction-to-intervention-n2aqewbwf) · Medium · 0 upvotes · 0 comments - [Prompt Engineering for Time Series Analysis](https://daily.dev/posts/prompt-engineering-for-time-series-analysis-zptrj3gxb) · Machine Learning Mastery · 4 upvotes · 0 comments - [Why Powerful ML Is Deceptively Easy — Part 2](https://daily.dev/posts/why-powerful-ml-is-deceptively-easy-part-2-mnfoiswv1) · Towards Data Science · 0 upvotes · 0 comments - [Product Experimentation with Regression-Based Causal Inference: Estimating LLM Feature Impact with Python and statsmodels](https://daily.dev/posts/product-experimentation-with-regression-based-causal-inference-estimating-llm-feature-impact-with-p-cgaks9y2k) · freeCodeCamp · 0 upvotes · 0 comments - [Data Bias: The Bias Behind Every Model](https://daily.dev/posts/data-bias-the-bias-behind-every-model-g4xlrbuof) · Medium · 0 upvotes · 0 comments --- Tags: [#machine-learning](https://daily.dev/tags/machine-learning), [#python](https://daily.dev/tags/python), [#feature-engineering](https://daily.dev/tags/feature-engineering) [View this post on daily.dev](https://daily.dev/posts/building-models-in-two-worlds-from-latent-constructs-to-behavioral-signals-uroobl2k7) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"Building Models in Two Worlds: From Latent Constructs to Behavioral Signals","url":"https://daily.dev/posts/building-models-in-two-worlds-from-latent-constructs-to-behavioral-signals-uroobl2k7","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/building-models-in-two-worlds-from-latent-constructs-to-behavioral-signals-uroobl2k7"},"datePublished":"2026-07-13T14:16:34.150Z","dateModified":"2026-09-14T07:12:45.951Z","description":"A data scientist compares modeling latent constructs (survey-based SEM in academic research) versus behavioral signals (ML-based propensity and lift models in...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/475db4f26964201a71ab7fdd0ba052bd?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/475db4f26964201a71ab7fdd0ba052bd?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"Towards Data Science","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Towards Data Science","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/tds","url":"https://daily.dev/sources/tds"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/building-models-in-two-worlds-from-latent-constructs-to-behavioral-signals-uroobl2k7","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"machine-learning,python,feature-engineering","timeRequired":"PT18M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Towards Data Science","item":"https://daily.dev/sources/tds"},{"@type":"ListItem","position":3,"name":"Building Models in Two Worlds: From Latent Constructs to Behavioral Signals"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/building-models-in-two-worlds-from-latent-constructs-to-behavioral-signals-uroobl2k7#faq","mainEntity":[{"@type":"Question","name":"Why did a geo lift test show that an advertising campaign caused negative sales for a snack-category product?","acceptedAnswer":{"@type":"Answer","text":"A seasonality mismatch between test and comparison regions caused the false negative reading, not an actual suppression effect from the ads. The product's sales ramp toward a seasonal peak at different times across regions, so during the campaign window the test region was still climbing while comparison regions had already crested, and after the campaign the test region read below comparison, mimicking an ad-caused decline. The real failure was the assumption that groups would have moved together absent the ad. Anyone debugging a suspicious lift or causal-inference result can find similar diagnostic breakdowns curated on daily.dev."}},{"@type":"Question","name":"Why is correlation among features good in a structural equation model but bad in a propensity or predictive model?","acceptedAnswer":{"@type":"Answer","text":"In structural equation modeling, correlated survey items within a construct are the evidence that they measure one underlying latent variable, so high internal correlation signals good reliability. In predictive models like propensity classifiers, that same redundancy among features destabilizes linear coefficients and smears feature importance across correlated tree splits, so it's treated as a problem rather than a virtue when interpretation matters. Developers weighing multicollinearity trade-offs across modeling approaches can track similar analysis on daily.dev."}},{"@type":"Question","name":"How do you validate a latent construct like privacy concern or perceived usefulness before using it in a regression model?","acceptedAnswer":{"@type":"Answer","text":"Build the construct from several survey indicators using structural equation modeling rather than averaging items directly, then check that items load properly on their intended construct, that the construct holds together internally via Cronbach's alpha or composite reliability, and that it explains more variance in its own indicators than it shares with other constructs via average variance extracted and discriminant validity. Averaging blindly lets measurement error bias coefficients toward zero. Practitioners building measurement models can find related methodology discussions on daily.dev."}}]} ```