Search

The academic and economic benefits of school spending are well-established, but focusing on these outcomes may underestimate the full social benefits of school spending. Recent increases in U.S. child mortality are driven by injuries and raise questions about what types of social investments could reduce child deaths. We use close school district tax elections and negative binomial regression models to estimate effects of a quasi-random increase in school spending on county child mortality. We find consistent evidence that increased school spending from passing a tax election reduces child mortality. Districts that narrowly passed a proposed tax increase spent an additional $243 per pupil, mostly on instruction and salaries, and had 4% lower child mortality after spending increased (6-10 years after the election). This increased spending also reduced child deaths of despair (due to drugs, alcohol, or suicide) by 5% and child deaths due to accidents or motor vehicle accidents by 7%. Estimates predicting potential mechanisms suggest that lower child mortality could partly reflect increases in the number of teachers and counselors, higher teacher salaries, and improved student engagement. [Funding for this report was provided by the Gilead Foundation Creating Possible Fund and the Population Studies and Training Center at Brown University.]

The EdResearch for Action "Design Principles Series" focuses on a single program or practice that has been proven to have positive effects on student outcomes. Authors -- leading experts from across the field of education research -- look across many high-quality studies of similar programs to identify the components and conditions that are key to its effective implementation. The "Design Principles Series" helps practitioners adapt and successfully implement an evidence-based program to meet the needs of their target population. This brief is an update of the previous version published in 2021. It incorporates new research on effective design and implementation components in high-impact tutoring programs. [This brief was prepared in collaboration with the University of Virginia, School of Education and Human Development and National Student Support Accelerator.]

In-person tutoring has been shown to improve academic achievement. Though less well-researched, virtual tutoring has also shown a positive effect on achievement but has only been studied in grade five or above. We present findings from the first randomized controlled trial of virtual tutoring for young children (grades K-2). Students were assigned to 1:1 tutoring, 2:1 tutoring, or a control group. Assignment to any virtual tutoring increased early literacy skills by 0.05-0.08 SD with the largest effects for 1:1 tutoring (0.07-0.12 SD). Students initially scoring well below benchmark and first graders experienced the largest gains from 1:1 tutoring (0.15 and 0.20 SD, respectively). Effects are smaller than typically seen from in-person early literacy tutoring programs but still positive and statistically significant, suggesting promise particularly in communities with in-person staffing challenges. [On Your Mark and Uplift Education were partners in this research and additional support was provided by the National Student Support Accelerator Team.]

College students make job decisions without complete information. As a result, they may rely on misleading heuristics ("interesting jobs pay badly") and pursue options misaligned with their goals. We test whether highlighting job characteristics changes decision making. We find increasing the salience of a job's monetary benefits increases the likelihood college students apply by 196%. In contrast, emphasizing prosocial, career, or social benefits has no effect, despite students identifying these benefits as primary motivators for applying. The study highlights the detrimental incongruencies in students' decision making alongside a simple strategy for recruiting college students to jobs that offer enriching experiences.

The clustering dipole in the 2MASS galaxy survey converges on a scale of ~400Mpc to the local peculiar velocity inferred from the Cosmic-Microwave-Background dipole. I show that this limits the graviton mass in Yukawa theories of gravity to less than 5x10^{-32}eV. The new limit is 2.5x10^8 times tighter than the latest constraint from gravitational waves detected by the LIGO-Virgo-KAGRA collaboration., Comment: Submitted to an AAS journal

Stellar mass black holes are formed from the terminal collapse of massive stars if the ensuing neutrino shock is unable to eject the stellar envelope. Direct observations of black hole formation remain inconclusive. We report observations of M31-2014-DS1, a massive, hydrogen-depleted supergiant in the Andromeda galaxy identified via a mid-infrared brightening in 2014. Its total luminosity remained nearly constant for the subsequent thousand days, before fading dramatically over the next thousand days by $\gtrsim 10\times$ and $\gtrsim 10^4\times$ in total and visible light, respectively. Together with the lack of a detected optical outburst, the observations are explained by the fallback of the stellar envelope into a newly formed black hole, moderated by the injection of a $\sim 10^{48}$ erg shock. Unifying these observations with a candidate in NGC 6946, we present a concordant picture for the birth of stellar mass black holes from stripped massive stars., Comment: Submitted for review

Generative AI, particularly Language Models (LMs), has the potential to transform real-world domains with societal impact, particularly where access to experts is limited. For example, in education, training novice educators with expert guidance is important for effectiveness but expensive, creating significant barriers to improving education quality at scale. This challenge disproportionately harms students from under-served communities, who stand to gain the most from high-quality education. We introduce Tutor CoPilot, a novel Human-AI approach that leverages a model of expert thinking to provide expert-like guidance to tutors as they tutor. This study is the first randomized controlled trial of a Human-AI system in live tutoring, involving 900 tutors and 1,800 K-12 students from historically under-served communities. Following a preregistered analysis plan, we find that students working with tutors that have access to Tutor CoPilot are 4 percentage points (p.p.) more likely to master topics (p<0.01). Notably, students of lower-rated tutors experienced the greatest benefit, improving mastery by 9 p.p. We find that Tutor CoPilot costs only $20 per-tutor annually. We analyze 550,000+ messages using classifiers to identify pedagogical strategies, and find that tutors with access to Tutor CoPilot are more likely to use high-quality strategies to foster student understanding (e.g., asking guiding questions) and less likely to give away the answer to the student. Tutor interviews highlight how Tutor CoPilot's guidance helps tutors to respond to student needs, though they flag issues in Tutor CoPilot, such as generating suggestions that are not grade-level appropriate. Altogether, our study of Tutor CoPilot demonstrates how Human-AI systems can scale expertise in real-world domains, bridge gaps in skills and create a future where high-quality education is accessible to all students., Comment: Our pre-registration for this randomized controlled trial can be found here: https://osf.io/8d6ha

We examine a century of radial velocity, visual magnitude, and astrometric observations of the nearest red supergiant, Betelgeuse, in order to reexamine the century-old assertion that Betelgeuse might be a spectroscopic binary. These data reveal Betelgeuse varying stochastically over years and decades due to its boiling, convective envelope, periodically with a $ 5.78$~yr long secondary period, and quasi-periodically from pulsations with periods of several hundred days. We show that the long secondary period is consistent between astrometric and RV datasets, and argue that it indicates a low-mass companion to Betelgeuse, less than a solar mass, orbiting in a 2,110 day period at a separation of just over twice Betelgeuse's radius. The companion star would be nearly twenty times less massive and a million times fainter than Betelgeuse, with similar effective temperature, effectively hiding it in plain sight near one of the best-studied stars in the night sky. The astrometric data favor an edge-on binary with orbital plane aligned with Betelgeuse's measured spin axis. Tidal spin-orbit interaction drains angular momentum from the orbit and spins up Betelgeuse, explaining the spin--orbit alignment and Betelgeuse's anomalously rapid spin. In the future, the orbit will decay until the companion is swallowed by Betelgeuse in the next 10,000 years., Comment: Submitted to AAS Journals, comments welcome

The Schwarzschild radii of primordial black holes (PBHs) in the mass range of 6x10^{14}g to 4x10^{19}g match the sizes of nuclei to atoms. I discuss the resulting quantum-mechanical suppression in the accretion of matter by PBHs in dense astrophysical environments, such as planets or stars., Comment: Accepted for publication in The Astrophysical Journal Letters, 7 pages

We use the inferred evolution of Lyman-$\alpha$ luminosities of galaxies in the redshift range of $z \sim 9-16$ from the $James \ Webb \ Space \ Telescope$ (JWST) data to predict the power spectrum of Loeb-Rybicki haloes formed by scattered Lyman-$\alpha$ radiation in the diffuse intergalactic hydrogen before reionization. We find excellent prospects for a statistical detection of the intensity mapping signal with current and future experiments such as the JWST, SPHEREx and CDIM. We also describe the detectability of the signal in cross-correlation with the 21-cm emission from atomic hydrogen in the intergalactic medium at these redshifts. We find that the cross-correlation signal should be detectable at a significance of a few tens of standard deviations out to $z \sim 16$ using the above experiments in combination with the Square Kilometre Array (SKA)-LOW and its pathfinder, the Murchison Widefield Array (MWA)., Comment: 8 pages, 5 figures, 2 tables

Current cosmological constraints allow primordial black holes (PBHs) to constitute dark matter in the mass range of $10^{18}$-$10^{22}$g. We show that a major portion of this logarithmic window can be ruled-out based on the Solar System ephemeris, given that the external mass enclosed within 50 au from the Sun did not change by more than $\sim 5\times 10^{-14}$ solar masses per year in recent decades., Comment: Accepted for publication in Research Notes of the American Astronomical Society (August 2024), 4 pages

We show that black holes can be made of light by adding gravity to the discussion of Alvarez-Dominguez et al., arXiv:2405.02389 [PRL 130, 041401 (2024)]., Comment: Comment on arXiv:2405.02389 [PRL 130, 041401 (2024)], 2 pages; submitted for publication as a comment to Physical Review Letters, with added references

The first interstellar objects, such as 'Oumuamua, Borisov, and IM1, were discovered over the past decade. We follow the trajectories of known interstellar objects in the gravitational potential of the Milky Way galaxy to constrain their possible origin. By initiating the trajectories based on their measured velocities relative to the Local Standard of Rest and considering their vertical excursion from the Milky Way's disk mid-plane, we estimate their likely ages. We find that 'Oumuamua likely originated near the mid-plane of the thin disk, suggesting a younger age. Comet Borisov's trajectory suggests an age similar to the Sun, while IM1's larger excursions suggest an older source. Finally, we project that human-made probes like Voyager 1 and Pioneer 10 will reach the opposite side of the Milky Way disk in approximately 2 Gyr and return to the Solar vicinity before the Sun becomes a red giant., Comment: 11 pages, 18 figures

The majority of low-mass ($\log_{10} M_*/M_{\odot}=9-10$) galaxies at high redshift ($z>1$) appear elongated in projection. We use JWST-CEERS observations to explore the role of gravitational lensing in this puzzle. The typical galaxy-galaxy lensing shear $\gamma\sim1\%$ is too low to explain the predominance of elongated early galaxies with ellipticity $e\approx0.6$. However, non-parametric quantile regression with Bayesian Additive Regression Trees reveals hints of an excess of tangentially-aligned source-lens pairs with $\gamma>10\%$. On larger scales, we also find evidence for weak lensing shear. We rule out the null hypothesis of randomly oriented galaxies at $\gtrsim99\%$ significance in multiple NIRCam chips, modules and pointings. The number of such regions is small and attributable to chance, but coherent alignment patterns suggest otherwise. On the chip scale, the average complex ellipticity $\langle e\rangle\sim10\%$ is non-negligible and beyond the level of our PSF uncertainties. The shear variance $\langle\overline{\gamma}^2\rangle\sim10^{-3}$ is an order of magnitude above the conventional weak lensing regime but is more sensitive to PSF systematics, intrinsic alignments, cosmic variance and other biases. Taking it as an upper limit, the maximum implied ``cosmic shear'' is only a few percent and cannot explain the elongated shapes of early galaxies. The alignments themselves may arise from lensing by a protocluster or filament at $z\sim0.75$ where we find an overabundance of massive lens galaxies. We recommend a weak lensing search for overdensities in ``blank'' deep fields with JWST and the Roman Space Telescope., Comment: Submitted to ApJ, main body is 26 pages with 18 figures, comments welcome

Recently, a new population of compact, high-redshift (z>7) galaxies appeared as little red dots (LRDs) in deep JWST observations. The latest spectroscopic data indicates that these galaxies contain an evolved stellar population, reflecting an early episode of high star-formation-rate. The appearance of broad emission lines suggests that a central overmassive black hole also powers these galaxies. I propose that LRD galaxies represent the low-spin tail of the galaxy population. Low-spin galaxies host a more compact gaseous disk with an enhanced star formation rate relative to typical galaxies at the same redshift. The compact disk feeds efficiently a central black hole, as predicted by Eisenstein & Loeb (1995)., Comment: Published in RNAAS, 4 pages

Dark matter particles were suggested to have an electric charge smaller than the elementary charge unit $e$. The behavior of such a medium is similar to a collisionless plasma. In this paper, we set new stringent constraints on the charge and mass of the millicharged dark matter particle based on observational data on the Bullet X-ray Cluster., Comment: (6 pages, 1 figure)

Recently, Pulsar Timing Arrays (PTAs) reported a signal at nanohertz frequencies consistent with a stochastic gravitational wave background. Here, I show that the Brownian motion of the Sun as a result of its random gravitational interactions with the cluster of thousands of unmodeled Main-belt asteroids of diameters <80km, combined with small inaccuracies in the estimated masses of larger asteroids, introduces timing noise for pulsars which is up to 20% of the reported signal at frequencies of ~1/(few-years). The asteroid contribution needs to be modeled better in order to obtain accurate inferences from the PTA signal., Comment: Accepted for publication in ApJL, 5 pages (final version)

The $James \ Webb \ Space \ Telescope \ (JWST)$ recently reported a large population of UV luminous galaxies at high redshifts, $ z > 10$, as well as Lyman-$\alpha$ emitting (LAE) galaxies out to $z \sim 11$. We use the observed UV luminosities along with a data-driven approach at lower redshifts to place constraints on the observability of the intergalactic Lyman-$\alpha$ intensity, scattered in the form of Loeb-Rybicki haloes, during the pre-reionization and reionization epochs ($z \sim 9-16$). We forecast the sensitivity and resolution required to detect these intergalactic haloes, finding that individual haloes with LAE luminosities $> 10^{43}$ ergs/s are detectable at a few sigma level at $z \lesssim 9$, while stacking of $\sim 10$ haloes is expected to result in detections out to $z \sim 16$. Finding these haloes is expected to shed light on the neutral intergalactic hydrogen during cosmic reionization., Comment: 15 pages, 3 figures, 2 tables; version accepted for publication in JCAP

We analyze consequences of trying to replace dark matter and dark energy with models of stochastic spacetime. In particular, we analyze the model put forth by Ref. [1], in which it is claimed that ``post-quantum classical gravity" (PQCG), a stochastic theory of gravity, leads to modified Newtonian dynamics (MOND) behavior on galactic scales that reproduces galactic rotation curves, and leads to dark energy. We show that this analysis has four basic problems: (i) the equations of PQCG do not lead to a new large scale force of the form claimed in the paper, (ii) the form claimed is not of the MONDian form anyhow and so does not correspond to observed galactic dynamics, (iii) the spectrum of fluctuations is very different from observations, and (iv) we also identify some theoretical problems in these models., Comment: 10 pages in double column format, 1 figure. V2: More details including analysis of fluctuations. V3: Further details. V4: Analysis of temporal correlations. V5: Updated towards version accepted for publication in JCAP

JWST is discovering a large population of $z>4$ supermassive black holes (SMBHs) that are overmassive with respect to the stellar content of their hosts. A previous study developed a physical model to interpret this overmassive population as the result of quasar feedback acting on a compact host galaxy. In this Note, we apply this model to JADES GN 1146115, a dormant supermassive black hole at $z=6.7$ whose mass is $\sim40\%$ of the host's mass in stars and accreting at $\sim2\%$ of the Eddington limit. The host has been forming stars at the low rate of $\sim 1 \, \rm M_\odot \,yr^{-1}$ for the past $\sim 100$ Myr. Our model suggests that this galactic system is on the verge of a resurgence of global star formation activity. This transition comes after a period of domination by the effect of its overmassive black hole, whose duration is comparable to typical quasar lifetimes., Comment: Accepted on Research Notes of the AAS. This is a concise note (3 pages, 1 figure) regarding the direct application of the model described in arXiv:2401.04159 to the source presented in arXiv:2403.03872

The strong tidal force in a supermassive black hole's (SMBH) vicinity, coupled with a higher stellar density at the center of a galaxy, make it an ideal location to study the interaction between stars and black holes. Two stars moving near the SMBH could collide at a very high speed, which can result in a high energy flare. The resulting debris can then accrete onto the SMBH, which could be observed as a separate event. We simulate the light curves resulting from the fallback accretion in the aftermath of a stellar collision near a SMBH. We investigate how it varies with physical parameters of the system. With all other physical parameters of the system held constant, the direction of the relative velocity vector at time of impact plays a large role in determining the overall form of the light curve. One distinctive light curve we notice is characterized by a sustained increase in the luminosity some time after accretion has started. We compare this form to the light curves of some candidate tidal disruption events (TDEs). Stellar collision accretion flares can take on unique appearances that would allow them to be easily distinguished, as well as elucidate underlying physical parameters of the system. There exist several ways to distinguish these events from TDEs, including the much wider range of SMBH masses stellar collisions may exist around.

In a recent preprint, Fernando et al. (2024) used public data from infrasound stations to constrain the localization of the fireball of the CNEOS 2014-01-08 (IM1) bolide. The analysis inferred a 90-percent-confidence ellipse with semi-minor and semi-major axes of 186 and 388 km, respectively. This large error ellipse includes the much better localization box derived by sensors aboard U.S. Government satellites which detected the fireball light. At the fireball's peak brightness, the CNEOS localization box documented by NASA/JPL measures 11.112km on a side and is centered on a latitude of 1.3S and a longitude of 147.6E. Here, we point out that the recent expedition to retrieve materials from IM1's site (Loeb et al. 2024a,b,c) surveyed a region of tens of km around the CNEOS box center, and was not dictated by the data studied by Fernando et al. (2024) because of its larger uncertainties., Comment: Submitted for publication. Reply to arXiv:2403.03966

Decisions about who is in charge of school systems often make the news. High-profile cases such as the recent takeover of the Houston Independent School District by the State of Texas -- where one of the first steps by state leaders was to replace the superintendent -- highlight the central importance of district leadership as well as the public perception of superintendents as critical contributors to the quality of a school system. Yet decision-makers who want to rely on research-based evidence about who to select or even how to conceptualize the characteristics required to be successful as superintendents have relatively little to draw on in the education research literature. In the spring of 2023, The Broad Center at the Yale School of Management partnered with the Annenberg Institute at Brown University to bring together a group of scholars who have made recent contributions to research on superintendency and to educational leadership more broadly. These conversations resulted in a strong consensus on the need for a renewed push to build knowledge around specific aspects of the superintendency, anchored in new data that broaden the research that is possible in the field. Future research, in particular, could inform the field by investigating superintendents' roles, including their time use, decision-making, priorities, management styles, and behaviors; the routes to the superintendency and factors associated with retention; how and why superintendents improve; and the causal impact of superintendents on key outcomes. [This report was co-produced by The Broad Center at Yale School of Management.]

We analyze the possibility of global anisotropy of the universe. We consider an altered Friedmann Lemaitre Robertson Walker metric in which there are different scale factors along the three different axes of space. We construct the corresponding altered Friedmann equations. We show that any initial anisotropies decrease into the future. At late times, the difference in Hubble parameters changes as $1/\sqrt{t}$ in a radiation dominated era and as $1/t$ in a matter dominated era. We use constraints from Big Bang Nucleosynthesis and the Cosmic Microwave Background to constrain the level of anisotropies at early times. We also examine how the approach back in time to the singularity is radically altered; happening much more abruptly, as a function of density, in an anisotropic universe. We also mention improved bounds that can arise from measurements of primordial gravitons, Weakly interacting massive particles, and neutrinos., Comment: 10 pages in double column format, 4 figures. V2: Added references. V3: Updated towards version published in Phys. Rev. D

Owing to the exceptional sensitivity of the Vera C. Rubin Observatory, we predict that its upcoming LSST images will be contaminated by numerous flares from centimeter-scale space debris in Low Earth Orbits (LEO). Millisecond-duration flares from these LEO objects are expected to produce detectable image streaks of a few arcseconds with AB magnitudes brighter than 14., Comment: 3 pages, submitted for publication in an AAS Journal

We have conducted an extensive towed-magnetic-sled survey during the period of June 14-28, 2023, over the seafloor about 85 km north of Manus Island, Papua New Guinea, centered around the calculated path of the bolide CNEOS 2014-01-08 (IM1). We found about 850 spherules of diameter 0.1-1.3 millimeters in our samples. The samples were analyzed by micro-XRF, Electron Probe Microanalyzer and ICP Mass spectrometry. Here we report major and trace element compositions of the samples and classify spherules based on that analysis. Spherules comprising 22% of the collection, appear to all reflect planetary igneous differentiation and are all different from previously described spherules. A subset of the differentiated-spherules show an excess of Be, La and U, by up to three orders of magnitude relative to the solar system standard of CI chondrites. Detailed mass spectrometry of 12 of these "BeLaU"-type spherules, the population of which may constitute up to ~10% of our entire collected sample, suggests that they are derived from material formed by planetary igneous fractionation. Their chemical composition is unlike any known solar system material. The "BeLaU"-type spherules reflect a highly differentiated, extremely evolved composition of an unknown source., Comment: Accepted for publication in Chemical Geology

JWST has detected many overmassive galactic systems at $z > 4$, where the mass of the black hole, $M_\bullet$, is $10-100$ times larger than expected from local relations, given the host's stellar mass, $M_\star$. This Letter presents a model to describe these overmassive systems in the high-$z$ Universe. We suggest that the black hole mass is the main driver of high-$z$ star formation quenching. SMBHs globally impact their high-$z$ galaxies because their hosts are physically small, and the black holes have duty cycles close to unity at $z > 4$. In this regime, we assume that black hole mass growth is regulated by the quasar's output, while stellar mass growth is quenched by it and uncorrelated to the global properties of the host halo. We find that the ratio $M_\bullet/M_\star$ controls the average star formation efficiency: if $M_\bullet/M_\star > 8\times 10^{18} (n \Lambda/f_{edd})[(\Omega_b M_h)/(\Omega_m M_\star) - 1]$, then the galaxy is unable to form stars efficiently. Once this ratio exceeds the threshold, a runaway process brings the originally overmassive system towards the local $M_\bullet - M_\star$ relation. Furthermore, the $M_\bullet - M_\star$ relation evolves with redshift as $\propto (1+z)^{5/2}$. At $z \sim 5$, we find an overmassive factor of $\sim 55$, in excellent agreement with current JWST data and the high-$z$ relation inferred from those. Extending the black hole horizon farther in redshift and lower in mass will test this model and improve our understanding of the early co-evolution of black holes and galaxies., Comment: Accepted for publication in The Astrophysical Journal. 11 pages, 4 figures

We use the recent statistics of dual active galactic nuclei (AGN) in the $James \ Webb \ Space \ Telescope$ (JWST) data at $z \sim 3.4$ to address two aspects of the feedback and evolution scenarios of supermassive black hole binaries (SMBHB). We find that the JWST data provide evidence for the members of a binary black hole being 'lit' at the same time, rather than independently -- a scenario which is consistent with gas-rich mergers being responsible for concurrent AGN activity. This conclusion is supported by the recent NANOGrav Pulsar Timing Array (PTA) measurements, whose upper limits on the stochastic gravitational wave strain amplitude lie below those expected from extrapolating the dual AGN fraction. The results indicate either a 'stalling' of the binaries at the separations probed by NANOGrav, or rapid gas-driven inspirals., Comment: 5 pages, 3 figures; references updated, typos corrected, matches version published in A&A Letters

Recently, Pandya et al. (2023) argued that the shapes of dwarf galaxies in JWST-CEERS observations show a prolate fraction that rises from ~25% at redshifts z=0.5-1 to ~50-80% at z=3-8. Here we suggest that this apparent change could result from a surface-brightness bias, favoring the detection of edge-on disks at low-luminosities and high-redshifts. Changing edge-on projections with an axis ratio of 10 to a face-on orientation reduces their apparent surface brightness by 2.5 magnitude per arcsec$^2$ and could shift a substantial fraction of the observed galaxies below the detection limit., Comment: 4 pages

We consider extremal problems related to decks and multidecks of rooted binary trees (a.k.a. rooted phylogenetic tree shapes). Here, the deck (resp. multideck) of a tree $T$ refers to the set (resp. multiset) of leaf induced binary subtrees of $T$. On the one hand, we consider the reconstruction of trees from their (multi)decks. We give lower and upper bounds on the minimum (multi)deck size required to uniquely encode a rooted binary tree on $n$ leaves. On the other hand, we consider problems related to deck cardinalities. In particular, we characterize trees with minimum-size as well as maximum-size decks. Finally, we present some exhaustive computations for $k$-universal trees, i.e., rooted binary trees that contain all $k$-leaf rooted binary trees as induced subtrees., Comment: 18 pages, 14 figures

Scaling high-quality tutoring remains a major challenge in education. Due to growing demand, many platforms employ novice tutors who, unlike experienced educators, struggle to address student mistakes and thus fail to seize prime learning opportunities. Our work explores the potential of large language models (LLMs) to close the novice-expert knowledge gap in remediating math mistakes. We contribute Bridge, a method that uses cognitive task analysis to translate an expert's latent thought process into a decision-making model for remediation. This involves an expert identifying (A) the student's error, (B) a remediation strategy, and (C) their intention before generating a response. We construct a dataset of 700 real tutoring conversations, annotated by experts with their decisions. We evaluate state-of-the-art LLMs on our dataset and find that the expert's decision-making model is critical for LLMs to close the gap: responses from GPT4 with expert decisions (e.g., "simplify the problem") are +76% more preferred than without. Additionally, context-sensitive decisions are critical to closing pedagogical gaps: random decisions decrease GPT4's response quality by -97% than expert decisions. Our work shows the potential of embedding expert thought processes in LLM generations to enhance their capability to bridge novice-expert knowledge gaps. Our dataset and code can be found at: \url{https://github.com/rosewang2008/bridge}., Comment: NAACL 2024. Code: https://github.com/rosewang2008/bridge