\section{Historical development of the literature}
\label{sec:history}

Hedonic regressions of housing prices predate \citet{rosen1974hedonic}: the method was used for price indices in the 1960s and
applied to housing by \citet{ridker1967determinants} for air pollution and by \citet{kain1970measuring} for the components of
housing quality. The 1970s and 1980s established the canonical specification, the problems of functional form
\citep{halvorsen1981choice,cropper1988choice} and the interpretation of dummy-variable coefficients in semi-logarithmic
equations \citep{halvorsen1980interpretation,kennedy1981estimation}, and the first reviews \citep{follain1985estimating,
freeman1979hedonic}. The 1990s brought spatial econometrics into the field \citep{can1992specification,dubin1992spatial,
basu1998analysis} and the recognition that the marketing process, the data source (transactions, listings, appraisals, rents) and
the submarket structure all affect estimated implicit prices. \citet{malpezzi2003hedonic} wrote the standard applied review of
this period.

The Sirmans programme belongs to the mid-2000s. \citet{sirmans2005composition} tabulated the characteristics used and the signs
and significance obtained in a large set of hedonic studies, and \citet{sirmans2006value} ran meta-regressions of the raw
coefficients of nine structural characteristics on location, time, data type and specification, finding that implicit prices
vary by location, that the data type matters for some characteristics, and that there was no clear change over time. The
programme was extended to time-on-market coefficients \citep{sirmans2010meta} and recently updated descriptively
\citep{khoshnoud2023composition}. Section~\ref{sec:questions} describes its limitations precisely. In parallel, attribute-specific
meta-analyses multiplied: transit proximity \citep{debrezion2007impact,mohammad2013meta,higgins2016forty}, airport and traffic
noise \citep{nelson2004meta}, open space and trees \citep{mcconnell2005value,waltert2010landscape,crompton2019impact}, water
quality and proximity \citep{guignet2022property,nicholls2018impact}, flood risk \citep{beltran2018is}, contamination and
hazards \citep{simons2006effect,braden2011meta}, energy efficiency \citep{cespedes2019meta}, school quality
\citep{nguyenhoang2011capitalization}, wind turbines \citep{parsons2022property} and crime \citep{macchiaroli2025crime}, each
pooling tens of studies with inverse-variance weights and, increasingly, publication-selection diagnostics
\citep{stanley2012meta}. The 2010s and 2020s added three developments that this synthesis must accommodate: quasi-experimental
designs for environmental and locational attributes; the systematic use of listing data from online platforms, with the
attendant shift from transaction to asking prices and from single-family houses to apartments in non-US markets; and machine
learning, first as a predictive competitor to hedonic regression \citep{wang2019machine,geerts2023machine} and then as a source of
new characteristics extracted from text and images.