🧪 This platform is in early beta. Features may change and you might encounter bugs. We appreciate your patience!
secular_trend_confounding
Secular trend confounding occurs when long-term background trends in both the exposure and outcome variables create the appearance of a causal relationship between them. If mobile phone ownership and diabetes prevalence both increase over two decades due to broader economic development, they will appear correlated even if mobile phones have no effect on diabetes.
An analysis of national data shows that organic food sales and autism diagnoses both rose dramatically from 2000 to 2015. Correlation = 0.99. Both trends are real, but both are independently driven by increased awareness, income, and screening — not by a causal link.
A wellness blogger plots smartphone ownership rates against childhood obesity rates from 2005 to 2020 and finds near-perfect correlation. Both trends are real, but both are independently driven by broad socioeconomic and lifestyle shifts over the same decade. Phones did not cause obesity — they simply rose in parallel.
A financial analyst reports that the number of women in the workforce correlates with rising household debt levels over a 30-year period. Both trends are genuine, but both are independently driven by long-term economic pressures, inflation, and changing social norms — not by any causal link between female employment and borrowing.
Binary (yes/no) questions an LLM must answer to identify this aspect:
Do both the exposure and outcome variables show long-term trends over the study period?
Type: binaryWere the trends adjusted for in the analysis (e.g., by detrending or including time as a covariate)?
Type: binaryCould changes in both variables be independently driven by broader societal or economic trends?
Type: binaryDoes the apparent relationship between exposure and outcome persist when the time trend is controlled?
Type: binarySecular trend confounding occurs when long-term background trends in both the exposure and outcome variables create the appearance of a causal relationship between them. If mobile phone ownership and diabetes prevalence both increase over two decades due to broader economic development, they will appear correlated even if mobile phones have no effect on diabetes.
Researchers often measure exposure and outcome at the same time points over years or decades. Shared secular trends create strong apparent associations that mimic causal effects and survive standard regression without explicit detrending.
Detrend time series data or include time as a covariate. Use interrupted time series analysis. Examine whether the correlation holds within subgroups with different trend trajectories.
Many ecological studies linking dietary patterns to disease outcomes are confounded by secular trends — both the dietary shift and the disease prevalence reflect broader changes in prosperity, urbanization, or diagnostic practices.
Use these tools to detect, analyze, or train this aspect.