research article on reading

Subscribe or renew today

Every print subscription comes with full digital access

Science News

It’s time to stop debating how to teach kids to read and follow the evidence.

Too many teachers are using the wrong approach

Many U.S. teachers are not using the most science-based approaches to teach reading.

Ariel Skelley/DigitalVision/Getty Images

Share this:

By Emily Sohn

April 26, 2020 at 7:00 am

On a chilly Tuesday back in January, my 7-year-old son’s classroom in Minneapolis was humming with reading activities. At their desks, first- and second-graders wrote on worksheets, read independently and did phonics lessons on iPads. In the hallway, students took turns playing a dice game that challenged them to spell out words with a consonant-vowel-consonant structure, like wig or map .

In another part of the classroom, small groups of two or three children, many missing their two front teeth, took turns sitting on a color-block carpet with teacher Patrice Pavek. In one group, Pavek asked students to read out loud from a list of words. “Con-fess,” said a dimpled 7-year-old named Hazel, who sat cross-legged in purple boots and a black fleece. Pavek reminded Hazel that a vowel sound in the middle of a word changes when you put an e at the end. Hazel tried again. “Con-fuse,” she said. “Beautiful!” Pavek beamed.

When Hazel returned to her desk, I asked her what goes through her mind when she gets to a word she doesn’t know. “Sound it out,” she said. “Or go to the next word.” Her classmates offered other tips. Reilly, age 6, said it helps to practice and look at pictures. Seven-year-old Beatrix, who loves books about unicorns and dragons, advocated looking at both pictures and letters. It feels weird when you don’t know a word, she said, because it seems like everyone else knows it. But learning to read is kind of fun, she added. “You can figure out a word you didn’t know before.”

Like the majority of schools in the United States, my son’s district uses an approach to reading instruction called balanced literacy. And that puts him and his classmates in the middle of a long-standing debate about how best to teach children to read.

The debate — often called the “reading wars” — is generally framed as a battle between two distinct views. On one side are those who advocate for an intensive emphasis on phonics: understanding the relationships between sounds and letters, with daily lessons that build on each other in a systematic order. On the other side are proponents of approaches that put a stronger emphasis on understanding meaning, with some sporadic phonics mixed in. Balanced literacy is one such example.

The issues are less black and white. Teachers and reading advocates argue about how much phonics to fit in, how it should be taught, and what other skills and instructional techniques matter, too. In various forms, the debate about how best to teach reading has stretched on for nearly two centuries, and along the way, it has picked up political, philosophical and emotional baggage.

In fact, science has a lot to say about reading and how to teach it. Plenty of evidence shows that children who receive systematic phonics instruction learn to read better and more rapidly than kids who don’t. But pitting phonics against other methods is an oversimplification of a complicated reality. Phonics is not the only kind of instruction that matters, and it is not the panacea that will solve the nation’s reading crisis.

Cutting through the confusion over how to teach reading is essential, experts say, because reading is crucial to success, and many people never learn to do it well.

According to U.S. government data, only one-third of fourth-graders have the reading skills to be considered proficient, which is defined by the National Assessment of Educational Progress as demonstrating competency over challenging subject matter. And a third of fourth-graders and more than a quarter of 12th-graders lack the reading skills to adequately complete grade-level schoolwork, says Timothy Shanahan, a reading researcher at the University of Illinois at Chicago.

Those struggles tend to persist. As many as 44 million U.S. adults, or 23 percent of the adult population, lack literacy skills , according to U.S. Department of Education data. Those affected may be able to read movie listings, or the time and place of a meeting, but they can’t synthesize information from long passages of text or decipher the warnings on medication inserts. People who can’t read well are less likely than others to vote, or read the news or secure employment. And today’s technology-based job market means students need to achieve more with reading than in the past, Shanahan says. “We are failing to do that.”

Trustworthy journalism comes at a price.

Scientists and journalists share a core belief in questioning, observing and verifying to reach the truth. Science News reports on crucial research and discovery across science disciplines. We need your financial support to make it happen – every contribution makes a difference.

Lessons in decoding

The vast majority of children need to be taught how to read. Even among those with no learning disabilities, only an estimated 5 percent figure out how to read with virtually no help, says Daniel Willingham, a psychologist at the University of Virginia in Charlottesville and author of Raising Kids Who Read . Yet educators have not reached consensus on how best to teach reading, and phonics is the part of the equation that people still argue about most.

The idea behind a systematic phonics approach is that children must learn how to translate the secret code of written language into the spoken language they know. This “decoding” begins with the development of phonological awareness, or the ability to distinguish between spoken sounds. Phonological awareness allows children, often beginning in preschool, to say that big and pig are different because of the sound at the beginning of the words.

Once children can hear the differences between sounds, phonics comes next, offering explicit instruction in the connections between letters, letter combinations and sounds. To be systematic, these skills need to be taught in an organized order of concepts that build on one another, preferably on a daily basis, says Louisa Moats, a licensed psychologist and literacy expert in Sun Valley, Idaho. Today, phonics proponents often advocate for the simple view of reading, which emphasizes decoding and comprehension, the ability to decipher meaning in sentences and passages.

Support for phonics has been around since at least the 1600s, but critics have also long expressed concerns that rote phonics lessons are boring, prevent kids from learning to love reading and distract from the ability to understand meaning in text. In the 1980s, this kind of thinking led to the rise of whole language, an approach aimed at making reading joyful and immersive instead of mindless and full of effort.

By the 2000s, a more all-around and phonics-inclusive approach called balanced literacy was gaining popularity as the leading theory in competition with phonics-first approaches.

In a 2019 survey of 674 early-elementary and special education teachers from around the United States, 72 percent said their schools use a balanced literacy approach , according to the Education Week Research Center, a nonprofit organization in Bethesda, Md. The implementation of balanced literacy, however, varies widely, especially in how much phonics is included, the survey found. That variation is probably preventing lots of kids from learning to read as well as they could, decades of research suggests.

In the late 1990s, with the reading wars in full swing, the National Institute of Child Health and Human Development brought together a panel of about a dozen reading experts to evaluate the evidence for how best to teach reading. The National Reading Panel’s first task was to figure out which types of teaching tasks to include in the analysis, says Shanahan, a panel member. Ultimately, the group chose eight categories and conducted a meta-analysis of 38 studies involving 66 controlled experiments from 1970 through 2000. The results showed support for five components of reading instruction that helped students the most.

Five essentials

A meta-analysis of 38 studies found five components of reading instruction were most helpful to students.

Phonemic awareness Knowing that spoken words are made of smaller segments of sound called phonemes

Phonics The knowledge that letters represent phonemes and that these sounds can combine to form words

Fluency The ability to read easily, accurately, quickly and with expression and understanding

Vocabulary Learning new words

Comprehension The ability to show understanding, often through summarization

Source: National Reading Panel

Two components that rose to the top were an emphasis on phonemic awareness (a part of phonological awareness that involves the ability to identify and manipulate individual sounds in spoken words) and phonics. Studies included in the analysis showed that higher levels of phonemic awareness in kindergarten and first grade were predictors of better reading skills later on. The analysis couldn’t assess the magnitude of benefits, but children who received systematic phonics instruction scored better on word reading, spelling and comprehension, especially when phonics lessons started before first grade. Those children were also better at sounding out words, including nonsense words, Shanahan says.

Vocabulary development was another essential component, as was a focus on comprehension. The final important facet was a focus on achieving fluency — the ability to read a text quickly, accurately and with proper expression — by having children read out loud, among other strategies.

Even before the panel released its results in 2000 , numerous studies and books from as early as the 1960s had concluded that there was value in explicit phonics instruction. Studies since then have added yet more support for phonics.

In 2008, the National Early Literacy Panel, a government-convened group that included Shanahan, considered dozens of studies on phonological awareness (including phonemic awareness) plus phonics instruction in preschool and kindergarten. Children who got decoding instruction scored substantially bette r on tests of phonological awareness compared with those who didn’t. The benefit was equivalent to a jump from the 50th percentile to the 79th percentile on standardized tests, suggesting those students were better prepared to learn how to read.

Likewise, a 2007 meta-analysis of 22 studies conducted in urban elementary schools found that minority children who received phonics instruction scored the equivalent of several months ahead of their minority peers on several academic measures. Studies have not addressed whether phonics might help close demographic achievement gaps, but research suggests that whole language approaches are less effective in disadvantaged populations than in other groups.

“There are several thousand studies at least that converge on this finding,” Moats says. “Phonics instruction has always had the edge in consensus reports.”

It is difficult to quantify how substantial the gains are from explicit phonics instruction, partly because the bulk of published research is full of ambiguities. Randomized trials are rare. Studies tend to be small. And in schools where teachers have autonomy to respond to students at their discretion, control groups are often not well-defined, making it hard to tell what phonics-focused programs are really being compared with, or how much phonics the control groups are getting. The reality of instruction can differ from classroom to classroom, even within the same school. And students who aren’t getting intensive phonics at school may have the blanks filled in at home, where parents might sound out words and talk about letters while reading bedtime stories.

The data that are available suggest that kids who get systematic phonics lessons score the equivalent of about half a grade level ahead of kids in other groups on standardized tests, Shanahan says. That’s not a giant leap, but it helps. “Overwhelmingly in studies, both individually and in a meta-analysis where you’re combining results across studies, if you explicitly teach phonics for some amount of time, kids do better than if you don’t pay much attention to that or if you pay a little bit of attention to [phonics],” he says.

Real experiences

Some of the most compelling evidence to support a phonics-focused approach comes from historical observations: When schools start teaching systematic phonics, test scores tend to go up. As phonics took hold in U.S. schools in the 1970s, fourth–graders began to do better on standardized reading tests.

In the 1980s, California replaced its phonics curriculum with a whole language approach. In 1994, the state’s fourth-graders tied for last place in the nation: Less than 18 percent had mastered reading. After California re-embraced phonics in the 1990s, test scores rose. By 2019, 32 percent achieved grade-level proficiency.

Those swings continue today. In 2019, Mississippi reported the nation’s largest improvement in reading scores ; the state had started training teachers in phonics instruction six years earlier. For the first time, Mississippi’s reading scores matched the nation’s average, with 32 percent of students showing proficiency, up from 22 percent in 2009, making it the only state to post significant gains in reading in 2019.

England, too, started seeing dramatic results after government-funded schools were required in 2006 to teach systematic phonics to 5- to 7-year-olds. When the country implemented a test to assess phonics skills in 2012, 58 percent of 5- and 6-year-olds passed. By 2016, 81 percent of students passed. Reading comprehension at age 7 has risen, and gains seem to persist at age 11. These population trends make a strong case for teaching phonics, says Douglas Fuchs, an educational psychologist at Vanderbilt University in Nashville.

A boost with phonics

After adding explicit phonics instruction statewide in 2013, Mississippi reported the nation’s largest improvement in reading scores among fourth-graders.

Mississippi fourth-grade reading proficiency

Source: National Assessment of Educational Progress 2019

Despite the evidence that children learn to read best when given systematic phonics along with other key components of a literacy program, many schools and teacher-training programs either ignore the science, apply it inconsistently or mix conflicting approaches that could hinder proficiency. In the 2019 Education Week Research Center survey, 86 percent of teachers who train teachers said they teach phonics. But surveyed elementary school teachers often use strategies that contradict a phonics-first approach: Seventy-five percent said they use a technique called three cuing. This method teaches children to guess words they don’t know by using context and picture clues, and has been criticized for getting in the way of learning to decode. More than half of the teachers said they thought students could understand written passages that contained unfamiliar words, even without a good grasp of phonics.

The disconnect starts at the top. In a 2013 review of nearly 700 teacher-training institutions, only 29 percent required teachers to take courses on four or five of the five essential facets of reading instruction identified by the National Reading Panel. Almost 60 percent required teachers to complete coursework on two or fewer of the essentials, according to the National Council on Teacher Quality, a research and policy group based in Washington, D.C.

Teacher’s choice

In a random sample of almost 700 U.S. early-elementary and special education teachers, most reported using a method called balanced literacy to teach reading. The simple view of reading, focused on phonics, was a distant second.

Balanced literacy Instruction includes a bit of everything, usually with some phonics.

Simple view The emphasis is on phonics, with a focus on two skills: decoding and language comprehension.

Whole language Instruction emphasizes whole words and phrases in meaningful contexts, including a strategy called three cuing.

How U.S. educators teach reading

Source: EdWeek Research Center 2020

In 2019, the Education Week Research Center also surveyed 533 postsecondary educators who train teachers on how to teach reading. Only 22 percent of those educators said their philosophy was to teach explicit, systematic phonics. Almost 60 percent said they support balanced literacy. And about 15 percent thought, contrary to evidence, that most students would learn to read if given the right books and enough time.

“The majority of classrooms in this country continue to embrace instructional practices and programs that do not include systematic instruction in foundational skills like phonemic awareness and phonics and spelling,” Moats says. “They just don’t do it.”

At my son’s Minneapolis school, reading specialist Karin Emerson told me about her early days teaching kindergarten, first and second grades in the 1990s. She was trained to use a whole language approach that included the three cuing technique.

Emerson described a typical reading lesson: “I’m going to show you a big book, and I’m going to cover up all of the letters of the word except the b , and I’m going to say, ‘Look at this page. It says this is a …’ What do you think it’s going to say?” Then she would point out the butterfly in the picture and ask the students to think about whether the b sound could refer to anything in the picture. “What does butterfly start with? A ‘b-uh.’ Do you think it’s going to be butterfly? I think it is going to be butterfly. It is.”

Eight years later, Emerson switched from classroom teacher to reading specialist, helping third-graders who weren’t reading yet. Many were the same students she had taught to read in younger grades. After reviewing the reading research, she implemented systematic phonics. By the end of third grade, students in her groups advanced an average of two grade levels. She now encourages early-grade teachers to add at least 20 minutes of phonics a day into literacy lessons.

Looking back to her classroom-teaching days, Emerson says parents often told her they were concerned that their children weren’t reading yet. “I would say, ‘Oh, they’ll be fine because they’re well spoken, they’re bright and you’re reading to them.’ Well they weren’t fine,” Emerson says. “Some people learn how to read super easy, and that’s great. But most people need to be taught, and there’s a pretty big chunk who need to be taught in a systematic way.”

While learning about ongoing battles over reading instruction, I have been marveling at my son’s transformation from nonreader to reader. One recent afternoon, he came home from school and told me that he had learned how to spell the word “A-G-A-I-N.” I asked him how he would spell it if it looked like it sounded. He worked it out, one sound at a time: “U-G-E-N.” We agreed the English language is pretty strange. It’s amazing anyone learns to read it at all.

This is your brain on reading

Reading is a relatively new activity for the human brain, which hasn’t had time to evolve specialized areas devoted to the task. Instead, our brains enlist areas, such as the visual system, that originated for other reasons, says Guinevere Eden, a neuroscientist at Georgetown University in Washington, D.C. An object like a tree or a lion needs to be recognizable from any angle, she says. But when we read, we need to override that kind of pattern recognition to distinguish, say, b from d , two letters that look identical to a beginning reader.

To translate squiggles and dots into sounds, several key brain areas, in both the visual and language systems, get involved. And how involved those areas are during reading shifts with increasing mastery, according to brain-imaging studies from the last two decades. When early or experienced readers sound out an unfamiliar word, they tap into the posterior and superior temporal lobes and inferior parietal lobe, which are involved in language and sensory processing. When the brain encounters a familiar word, on the other hand, the visual cortex takes over, suggesting that known words become like any other object that the brain recognizes instantly. As a person’s reading skills improve and the mental menu of familiar words grows, activity is more pronounced in the visual cortex during reading, Eden says.

Eden uses brain scans to understand what goes wrong in children with reading disabilities, who have trouble sounding out words. One of her goals is to evaluate interventions for children with dyslexia to see if the interventions target the brain processes that are most impaired.

Despite heavy marketing by companies that sell reading products using brain scans as evidence that the companies’ methods help children learn to read, Eden says that imaging studies cannot yet answer questions about which types of reading instruction are best for children, with or without reading disabilities.

More Stories from Science News on Humans

Bird flu can infect cats. What does that mean for their people?

Privacy remains an issue with several women’s health apps

Malnutrition’s effects on the body don’t end when food arrives

Biological puzzles abound in an up-close look at a human brain

Two distinct neural pathways may make opioids like fentanyl so addictive

Young people’s use of diabetes and weight loss drugs is up 600 percent 

Human body lice could harbor the plague and spread it through biting 

One of the world’s earliest farming villages housed surprisingly few people

Subscribers, enter your e-mail address for full access to the Science News archives and digital editions.

Not a subscriber? Become one now .

An official website of the United States government

The .gov means it’s official. Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

The site is secure. The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

• Publications
• Account settings

Preview improvements coming to the PMC website in October 2024. Learn More or Try it out now .

• Advanced Search
• Journal List
• HHS Author Manuscripts

The Effects of Reading Fluency Interventions on the Reading Fluency and Reading Comprehension Performance of Elementary Students With Learning Disabilities: A Synthesis of the Research from 2001 to 2014

Elizabeth a. stevens.

1 University of Texas at Austin, USA

Melodee A. Walker

Sharon vaughn.

Fluent word reading is hypothesized to facilitate reading comprehension by improving automatic word reading, thus releasing a reader’s cognitive resources to focus on meaning. Many students with learning disabilities (LD) struggle to develop reading fluency, which affects reading comprehension. This synthesis extends Chard, Vaughn, and Tyler’s (2002) review, synthesizing fluency intervention research from 2001 to 2014. The search yielded 19 studies examining reading fluency and comprehension outcomes of reading fluency interventions for students with LD in kindergarten through 5th grade. Results showed repeated reading (RR), multicomponent interventions, and assisted reading with audiobooks produced gains in reading fluency and comprehension. Providing a model of fluent reading and performance feedback, using easier level text, setting a performance criterion, and practicing RR with peers also contributed to improved outcomes. Findings suggest that RR remains the most effective intervention for improving reading fluency for students with LD. Limitations include sample size, only three group design studies, and infrequent use of standardized measures.

Fast and accurate word reading is hypothesized to facilitate reading comprehension because it releases a reader’s cognitive resources (e.g., working memory) to focus on meaning ( LaBerge & Samuels, 1974 ; Perfetti, 1980 , 1985 ; Wolf & Katzir-Cohen, 2001 ). When word recognition is slow and labored, cognitive load is occupied at the expense of understanding text. Due to the connection between efficient reading of connected text and comprehension, researchers in the past several decades have highlighted the importance of fluency instruction ( Bashir & Hook, 2009 ; Chard, Vaughn, & Tyler, 2002 ; Therrien, 2004 ; Welsch, 2006 ). In particular, students with learning disabilities (LD) struggle to develop reading fluency ( Bashir & Hook, 2009 ; Chard et al., 2002 ; Chard, Ketterlin-Geller, Baker, Doabler, & Apichatabutra, 2009 ). Reading can become a frustrating experience, which leads to an aversion to reading tasks. Consequently, students with LD may spend less time actually reading than proficient readers. When students with LD spend less time with text, this negatively affects vocabulary acquisition and comprehension development and may ultimately further contribute to the achievement gap for this population ( L. S. Fuchs, Fuchs, & Compton, 2010 ). As such, fluency interventions are integral to effective reading instruction for students with reading difficulties ( Bashir & Hook, 2009 ; Chard et al., 2009 ; Morgan, Sideridis, & Hua, 2012 ; Therrien, 2004 ).

The National Reading Panel ([NRP] 2000) identified fluency as one of the critical factors necessary for reading comprehension, but findings from observation studies examining the components of reading taught indicated fluency instruction is often overlooked for students with LD (e.g., Swanson, 2008 ). Guided oral repeated reading (RR) with teacher or peer feedback was identified as an effective method for improving reading fluency and comprehension for all readers. Recommendations from NRP were incorporated into the No Child Left Behind Act (2002) and the reauthorization of the Individuals With Disabilities Education Act (2004), requiring schools to assess individual progress, use research-based practices, and meet the needs of students with LD via specialized instruction, including in the area of reading fluency.

A systematic review of effective fluency interventions between 1975 and 2000 for elementary-age students with LD was conducted by Chard et al. (2002) . Findings indicated that RR interventions improved reading rate, accuracy, and comprehension. Additionally, RR with a model (e.g., teacher, computer, audio recording) was more effective than RR without a model, and modeling of fluent reading improved comprehension. Rereading multiple times, providing error correction feedback, and using progressively more difficult text were also associated with improved reading fluency. Findings related to RR with a peer were unclear. Results showed reading with a peer was an ineffective intervention by itself, yielding negative to small effect sizes on measures of oral reading fluency and comprehension ( Deno, Diment, Dongil, Marston, & Rogers, 1995 ; Mathes & Fuchs, 1993 ). When combined with other intervention components, such as partner retelling, summarizing, and predicting, peer RR yielded moderate to large effect sizes ( D. Fuchs, Fuchs, Mathes, & Simmons, 1997 ; Simmons, Fuchs, Fuchs, Mathes, & Hodge, 1995 ).

Despite legislation and recent research that promote best practices in reading fluency instruction, the National Assessment of Educational Progress (NAEP; National Center for Education Statistics, 2013 ) indicates that only 35% of fourth-grade students and 11% of students with LD are performing at a proficient level in reading. Vaughn and Wanzek (2014) described the concerning trend in the NAEP scores from 2002 to 2011: “While students without disabilities are improving their reading performance, the performance of students with disabilities is declining” (p. 47). This national trend suggests that educators are not meeting the needs of students with LD to close the gap with typically achieving peers in reading. In addition to empirical evidence, observation studies have revealed a discrepancy between best practices and instruction. For example, Vaughn and Wanzek reviewed observational research findings since 2000 to assess the quality of reading instruction for students with LD. Results indicated that students with LD are passively learning in large group settings with little to no specialized, intensive, and explicit instruction. Furthermore, many teachers believe students will naturally acquire fluent reading behaviors through sustained silent reading; however, struggling readers require direct instruction in reading fluency in order to make gains in fluency and comprehension ( Rasinski, Homan, & Biggs, 2009 ).

Researchers have continued to investigate the effects of fluency interventions since Chard et al.’s (2002) synthesis. For example, Therrien’s (2004) meta-analysis evaluated the effects of RR on familiar and unfamiliar (i.e., transfer) passages for students with LD in kindergarten through 12th grade. Therrien suggested that RR improves fluency and comprehension of familiar texts and may improve fluency and comprehension on transfer tasks. Similarly, Morgan and Sideridis (2006) synthesized alternative fluency interventions in single-subject-design studies. Interventions that provided vocabulary definitions and listening passage preview, goal setting and performance feedback, listening preview and repeated practice, and peer tutoring showed promise for increasing students’ reading fluency. Strickland, Boon, and Spencer (2013) reviewed RR research on elementary-age students with LD between 2001 and 2011; however, data for students with LD was not disaggregated for all of the studies included.

The most recent synthesis that focuses explicitly on reading fluency in elementary students with LD was Chard et al.’s (2002) synthesis. Since this synthesis was published, the Institute of Education Sciences’ (2011) What Works Clearinghouse (WWC) was established to identify high-quality research and provide guidance to educational stake-holders regarding research-based practices. WWC also has as its focus describing standards for high-quality research. This proposed synthesis aims to identify further high-quality knowledge regarding reading fluency interventions. It remains unclear which interventions, including technology-based fluency interventions, produce the best outcomes in fluency and comprehension for elementary students with LD within the framework of WWC standards. Given the declining reading performance of elementary students with LD, the purpose of this systematic review is to synthesize fluency intervention studies for this population published since 2001. The following research question was addressed: Which fluency interventions are associated with positive outcomes in reading fluency and comprehension for students with LD in kindergarten through 5th grade?

Operational Definitions

In this review, learning disability refers to an educationally identified or psychologist-diagnosed LD. Some studies specify the criteria used for identifying an LD; however, other studies do not include this information. Reading fluency is defined as the ability to read with speed, accuracy, and appropriate expression (NRP, 2000). Fluency intervention refers to any intervention that addresses students’ speed, accuracy, and prosody when reading text. Repeated reading is fluency practice in which a student repeatedly reads a passage aloud to increase oral reading fluency. The number of rereadings, level of text, type of performance feedback, and performance criterion may vary.

Search Procedures

This systematic review extends the corpus of studies identified by Chard et al. (2002) ; thus, there are many similarities in the search and inclusion procedures. The differentiating procedures of the present review are described below. In accordance with the procedures used in the Chard et al. (2002) study, an initial computer search was conducted of four electronic databases: Educational Resources Information Clearinghouse, PsycINFO, Education Source, and Academic Search Complete. Education Source was an additional database not utilized in Chard et al. (2002) , and Academic Search replaced ArticleFirst, which was no longer available. The search was limited to studies in peer-reviewed journals published between January 2001 and September 2014. The search terms used were disabilit* OR disorder, read* , and reading fluency, fluency, reading aloud, reading rate, repeated reading, reading practice, assisted reading, oral reading, paired reading, rereading, reading speed, reading expression, reading prosody, reading accuracy, and partner reading. The search yielded 4,135 articles; the abstracts of these studies were reviewed to identify studies that met the inclusion criteria. An ancestral search was completed using the reference lists from relevant syntheses and meta-analyses conducted since 2002 ( Chard et al., 2009 ; Morgan & Sideridis, 2006 ; Morgan et al., 2012 ; Strickland et al., 2013 ; Therrien, 2004 ). Hand searches were conducted in the same journals as those searched in the Chard et al. (2002) study, including Annals of Dyslexia, Education and Treatment of Children; Exceptional Children; Journal of Educational Research; Journal of Experimental Psychology General; Journal of Learning Disabilities; Journal of Literacy Research (formerly Journal of Reading Behavior ); Journal of Experimental Psychology: Learning, Memory, and Cognition; Learning Disabilities Quarterly; Learning Disabilities Research and Practice; Psychology in the Schools; Reading Horizons; Literacy Research and Instruction (formerly Reading Research and Instruction ); Reading Research Quarterly; Remedial Special Education; and School Psychology Review. Finally, reference lists of the studies that met the criteria were reviewed to identify any additional articles for inclusion.

Selection Criteria

Studies were included if they met four inclusion criteria. First, the participants in the study had to include students identified with LD in grades K through 5. Studies with combined samples of students with and without LD were included if the data for students with LD were disaggregated. Studies with participants who were at risk for reading failure or identified as having reading difficulties were excluded. As for grade level, participants had to be in grades K through 5 to be included. Studies with students older than fifth grade were included if 50% or more of the sample fell within the target grade range. Second, the intervention being implemented had to target reading fluency among connected text in English. Studies that included a combined intervention package were included if 50% or more of the intervention was spent on reading fluency. Studies targeting reading fluency in other languages or those that occurred outside of the school programming, such as in home, clinic, or camp settings, were excluded. Third, the studies had to employ an experimental, quasi-experimental, or single-subject design providing a treatment and comparison to determine experimental effect. Studies were excluded if they used single-group (pretest/ posttest), AB single-subject, descriptive, case study, or qualitative designs. Last, the dependent variable had to address outcomes in either reading fluency (rate, accuracy, and/or prosody) or comprehension.

Coding Procedures

Studies that met the inclusion criteria were coded using a coding protocol developed for education-related intervention research ( Vaughn, Elbaum, Wanzek, Scammacca, & Walker, 2014 ). The following data were extracted from each study: (a) participant information (e.g., age, grade level, number of participants with LD), (b) research design, (c) treatment fidelity, (d) description of treatment and comparison group(s), (e) clarity of causal inference, (f) measures, and (g) results and effect sizes. Coders participated in a 4.5-hour training session in which the protocol was described and coding procedures were applied to several sample studies of different design types. All studies were double-coded by the first author and an experienced graduate student coder; 99% interrater agreement was achieved, and discrepancies in coding were resolved via discussion.

After sorting the abstracts, 70 articles were further reviewed to determine if they met criteria; 3 studies were excluded based on single-group design ( Burns, Dean, & Foley, 2004 ; Stebbins, Stormont, Lembke, Wilson, & Clippard, 2012 ; Therrien & Kubina Jr., 2007 ). Forty-nine studies were excluded based on language, intervention components (i.e., decoding emphasis rather than fluency in connected text), setting, or participant characteristics. Nineteen studies that met inclusion criteria are organized into four tables based on features of the intervention (i.e., RR with or without a model, RR with multiple features, and interventions other than RR).

For group design studies, effect sizes (ESs) were calculated as the difference between the groups’ means divided by the pooled standard deviation; ES is interpreted using the following criteria: 0.8 is large, 0.5 is moderate, and 0.2 is small ( Cohen, 1988 ). Hedge’s g is reported to provide a less biased estimate of ES with particularly small samples ( Hedges, 1985 ).

For single-subject-design studies, data are reported as gains in words correct per minute (WCPM; i.e., the baseline phase or preintervention mean subtracted from the treatment phase mean) and decreases in errors per minute (EPM; i.e., the treatment phase mean subtracted from the baseline or preintervention mean) for each student. However, if mean performance data were unavailable across phases, the percentage of nonoverlapping data points (PND) was calculated using the graphs provided. For WCPM, PND was calculated by counting the total number of data points during the intervention phase that exceeded the highest baseline data point, divided by the total number of treatment data points and multiplying by 100 ( Scruggs, Mastropieri, & Casto, 1987 ). For EPM, PND was calculated by counting the number of data points during the intervention phase that did not exceed the lowest baseline data point, divided by the total number of treatment data points and multiplying by 100. PND results are interpreted as follows: 90% or greater is highly effective, 70% to 90% is moderately effective, 50% to 70% is minimally effective, and 50% or less is ineffective ( Scruggs & Mastropieri, 1998 ).

RR Without a Model

Table 1 summarizes five studies that examined the effects of repeatedly reading text, ranging two to four times, without modeling by a more proficient reader. O’Connor, White, and Swanson (2007) compared the effects of three RRs to 15 min of continuous reading and a no-treatment comparison. Results showed medium to large effects in favor of RR compared to the no-treatment comparison on standardized measures of reading fluency ( Gray Oral Reading Test–4 g = 0.65; Woodcock Reading Mastery Test–NU g = 0.94) and passage comprehension ( g = 0.72; g = 2.09); RR outper-formed continuous reading with small to medium effects in reading fluency ( g = 0.50; g = 0.29) and comprehension ( g = 0.59; g = 0.28).

Studies Examining Repeated Reading Without a Model.

Note. LD = learning disability; NR = not reported; RR = repeated reading; WCPM = words correct per minute; EPM = errors per minute; GORT-4 = Gray Oral Reading Test—4; WRMT-NU = Woodcock Reading Mastery Test-NU; PND = percentage of nonoverlapping data points; Gen. = generalization; HCO = high content overlap; M gain = the baseline phase or pre-intervention mean subtracted from the treatment phase mean; M decrease = the treatment phase mean subtracted from the baseline or pre-intervention phase mean; S1 = Student 1; S2 = Student 2; S3 = Student 3; S4 = Student 4; SE = standard error.

Three alternating treatment designs yielded positive results for improving rate, accuracy, and comprehension. Chafouleas, Martens, Dobson, Weinstein, and Gardner (2004) found three RR conditions (i.e., RR alone, RR with WCPM performance feedback (PF) and RR/PF with a reward option) to be highly effective for improving reading rate but minimally to moderately effective for improving accuracy. Kubina, Amato, Schwilk, and Therrien (2008) found RR to a high-performance criterion required more practice sessions but yielded greater gains in reading rate than RR to a low-performance criterion. However, maintenance testing showed decrement in WCPM regardless of a high or low performance standard. Finally, Welsch (2007) compared RR to a listening passage preview (LPP) condition of instructional and easier level texts; RR of instructional ( n = 1) and easier ( n = 3) text was identified as the best treatment condition (i.e., produced the best oral reading fluency rates during brief and extended analysis of alternating conditions). PND results showed RR to be highly effective for improving reading rate, accuracy, and comprehension.

Nelson, Alber, and Gordy (2004) compared systematic error correction (EC) alone to EC with RR on unfamiliar and previously read basal passages. The EC with RR conditions outperformed the EC-only condition; however, EC with RR of previously read material yielded better reading rate and accuracy than EC with RR of new text.

Generalization to novel passages

Only two studies examined generalization of skill to unfamiliar passages, and the results are inconclusive. In Chafouleas et al. (2004) , PND of student performance on new passages with high content overlap suggests generalization of rate and accuracy skills; however, further inspection of the graphs shows limited generalization of skills. For three of four students ( Welsch, 2007 ), RR yielded results that were moderately to highly effective in generalization (i.e., rates and comprehension to unfamiliar text at the same readability level). However, RR was ineffective for decreasing error rates in generalization passages for two students and moderately effective for the other two students.

RR With a Model

Nine studies examining RR with a model are grouped in Table 2 based on the type of modeling provided.

Studies Examining Repeated Reading With a Model.

Note. wwc = meets What Works Clearinghouse group design standards without reservations; Exp. = experiment; ELL = English language learner; LD = learning disability; WCPM = words correct per minute; EPM = errors per minute; NR = not reported; CRAB = Comprehensive Reading Assessment battery; HCO = high content overlap; PND = percentage of nonoverlapping data points; DORF = Dynamic Indicators of Early Literacy Skills Oral Reading Fluency; 1 = Experiment 1; 3 = Experiment 3; S1 = Student 1; S2 = Student 2; S3 = Student 3; SE = standard error.

Modeling by an adult

Four single-subject-design studies incorporated adult modeling of fluent reading ( Ardoin, Williams, Klubnik, & McCall, 2009 ; Daly, Bonfiglio, Mattson, Persampieri, & Foreman-Yates, 2005 ; Hapstak & Tracey, 2007 ; Welsch, 2007 ). Welsch’s (2007) alternating treatment design, listed in Table 1 (i.e., two of the four alternating treatments included RR without a teacher model), determined LPP without RR was not a best treatment condition compared to RR alone. However, three studies found that adult modeling combined with RR improved reading rates ( Ardoin et al., 2009 ; Daly et al., 2005 ; Hapstak & Tracey, 2007 ). Modeling and RR of instructional level text six versus three times yielded higher reading rates ( Ardoin et al., 2009 ). Engaging in echo or choral reading with a fluent adult model and practicing phrase drill EC improved reading rates on familiar passages ( Hapstak & Tracey, 2007 ).

Modeling and RR six versus three times did not provide greater generalization in reading rate to new, high-content overlap passages ( Ardoin et al., 2009 ). Adult modeling, RR, and phrase drill EC increased reading rates and decreased error rates on easy and difficult generalization passages with high content overlap ( Daly et al., 2005 ).

Modeling by a more proficient peer

Three studies examined modeling by a more proficient peer, which yielded favorable results for improving reading rate and comprehension but were ineffective for improving accuracy. Sáenz, Fuchs, and Fuchs (2005) compared Peer Assisted Learning Strategies to a traditional reading program for English-language learners also identified with LD; results yielded small effect sizes on a standardized measure of reading rate ( g = 0.42) and a maze comprehension task ( g = 0.46) but a large effect size for answering comprehension questions ( g = 0.91). Decker and Buggey (2014) used echo reading with an adult to make a video recording of each student’s reading. Excerpts of the student’s reading were spliced together to present a complete model of fluent reading. Video self-modeling (VSM) and video peer modeling outperformed the comparison group in mean gain of words read correctly per minute; however, VSM outperformed peer modeling in reading rate during intervention and maintenance, more than doubling the mean gain in WCPM. Oddo, Barnett, Hawkins, and Musti-Rao (2010) found peer RR in mixed-ability groups to be moderately effective for improving reading rate and comprehension on standardized passages but ineffective for improving reading accuracy.

Modeling by a matched-reading-ability peer

Four multiple-base-line studies examined peer RR with matched-reading-ability pairs. RR among struggling reader pairs showed small gains in WCPM on standardized passages ( Musti-Rao, Hawkins, & Barkley, 2009 ). Staubitz, Cartledge, Yurick, and Lo (2005) found that RR with a matched-ability peer increased reading rate, accuracy, and comprehension on below-grade-level children’s books for one student with LD; however, the student was unable to progress to grade level passages. Finally, Yurick, Robinson, Cartledge, Lo, and Evans (2006) reported three studies, two of which met inclusion criteria, examining peer RR among matched-reading-ability pairs. Unlike Staubitz et al. (2005) , Experiments 1 and 3 showed gains in reading rate using below-, on-, and above-grade-level passages. Experiment 1 also showed gains in comprehension and small gains in accuracy for the 2 fifth-grade students on fourth-, fifth-, and sixth-grade passages.

Staubitz et al. (2005) examined generalization results in three conditions (i.e., covert timing, overt timing, and overt timing with charting performance). The overt timing condition yielded the best reading rate, when the student was aware of being timed, but comprehension and accuracy decreased. Though the covert condition (i.e., the student was unaware of being timed) yielded a lower reading rate, it provided better accuracy and comprehension gains. Experiment 3 of Yurick et al. (2006) also assessed generalization under the same conditions. Results were inconsistent; the timed and charted condition provided the greatest increase in words per minute, but all three conditions resulted in decreased accuracy. Each condition showed similar, yet minimal, gains in comprehension.

RR Interventions With Multiple Features

Table 3 includes one group design and two multiple-base-line studies that examined RR as part of a multicomponent intervention (e.g., combined with vocabulary or comprehension instruction). In Sáenz et al. (2005) Peer Assisted Learning Strategies (i.e., one RR of text with EC, summarizing, and making predictions) outperformed the no-treatment comparison group with a large effect size on a standardized measure of comprehension questions answered correctly ( g = 0.91) and small effect sizes on standardized measures of reading accuracy ( g = 0.42) and a comprehension maze task ( g = 0.46).

Studies Examining Repeated Reading With Multiple Features.

Note. wwc = meets What Works Clearinghouse group design standards without reservations; LD = learning disability; ELL = English-language learner; RR = repeated reading; NR = not reported; WCPM = words correct per minute; CRAB = Comprehensive Reading Assessment Battery; EPM = errors per minute; SI = Student 1; S2 = Student 2; PND = percentage of nonoverlapping data points; SE = standard error.

Tam, Heward, and Heng (2006) compared RR of a familiar passage across sessions until a performance criterion was achieved to RR of a new passage each session without a criterion. Both conditions also included vocabulary instruction and EC. Students experienced similar gains in fluency, accuracy, and comprehension in both conditions. Hitchcock, Prater, and Dowrick (2004) compared tutoring in reading fluency (TRF; i.e., echo reading to a performance criterion), tutoring in reading fluency with VSM (TRF/VSM), tutoring in reading comprehension (i.e., fluency practice activities combined with direct instruction in story elements; Tutoring Reading Comprehension [TRC]), and tutoring in reading comprehension with VSM (TRC/VSM). TRF and TRF/VSM yielded ineffective to moderately effective results for improving WCPM, whereas TRC and TRC/VSM yielded highly effective results (i.e., 100% PND). Even though the results seem to favor TRC, it may be that students’ reading rates progressively increased in each phase with continued choral and echo reading practice.

Both conditions in Tam et al. (2006) produced minimal generalization of comprehension skill to unfamiliar passages; however, the vocabulary instruction with RR of the same passage to a performance criterion yielded greater generalization results than RR of a new passage each session ( Tam et al., 2006 ). Generalization results in Hitchcock et al. (2004) were reported at 100% PND; however, it is unclear to which condition this is attributed.

Studies Examining Fluency Interventions Other Than RR

Table 4 includes two studies that examined fluency interventions other than RR. Esteves and Whitten (2011) found assisted audiobook reading produced greater results on a standardized measure of oral reading fluency than sustained silent reading ( g = 1.07). Watson, Fore, and Boon (2009) compared word-supply and phonics-based feedback conditions, with the word-supply condition yielding a higher mean gain in WCPM. RR was not a focus of either intervention, but the same passage was used in each condition across five sessions. Thus, results may be more representative of the effects of RR combined with word-supply or phonics-based feedback.

Studies Examining Fluency Interventions Other Than Repeated Reading.

Note. NR = not reported; LD = learning disability; DORF = Dynamic Indicators of Early Literacy Skills Oral Reading Fluency; WCPM = words correct per minute; SE = standard error; S1 = Student 1.

Studies That Examined Types and Levels of PF in RR

PF during RR ranged from stating the number of correct words or errors per minute, to supplying the correct word, to providing syllable segmentation and blending instruction. Providing minimal feedback, such as reporting a student’s WCPM, may increase the reading rate when combined with RR (i.e., according to visual inspection of the student’s graph; Chafouleas et al., 2004 ). Word supply feedback improved reading rates when combined with RR but may not improve accuracy ( Nelson et al., 2004 ). Word supply with RR produced higher reading rates than phonics-based feedback with RR (i.e., sounding out miscues; Watson et al., 2009 ). RR with phrase drill EC and syllable segmentation and blending increased WCPM and decreased EPM ( Daly et al., 2005 ). Finally, incorporating EC feedback during paired RR yielded gains in reading rate for below-, on-, and above-grade-level texts ( Yurick et al., 2006 ).

Researchers have investigated the effects of fluency interventions since the release of the NRP report ( Morgan & Sideridis, 2006 ; Strickland et al., 2013 ; Therrien, 2004 ). These syntheses were limited by design type (single subject only) or intervention (RR); furthermore, the Strickland et al. (2013) synthesis did not disaggregate data for students with LD in all studies included. The Chard et al. (2002) synthesis is the most recent synthesis of fluency interventions, including those other than RR, that focused explicitly on students with LD in Grades K through 5. The review extends the Chard et al. synthesis on fluency interventions for students with LD between January 2001 and September 2014. Given the higher research standards set forth by WWC, the goal was to identify which fluency interventions produce the most improved outcomes in reading fluency and comprehension for this population. Unfortunately, only one study met WWC group design standards without reservations ( Sáenz et al., 2005 ). Five studies were ineligible for review because they did not use a sample aligned with the review protocol (i.e., at least 50% of the students not identified with LD; Ardoin et al., 2009 ; Hapstak & Tracey, 2007 ; Musti-Rao et al., 2009 ; O’Connor et al., 2007 ; Oddo et al., 2010 ). Welsch (2007) did not meet WWC pilot single-case design standards due to interassessor agreement. WWC has not reviewed the remaining 11 studies ( Chafouleas et al., 2004 ; Daly et al., 2005 ; Decker & Buggey, 2014 ; Esteves & Whitten, 2011 ; Hitchcock et al., 2004 ; Kubina et al., 2008 ; Nelson et al., 2004 ; Staubitz et al., 2005 ; Tam et al., 2006 ; Watson et al., 2009 ; Yurick et al., 2006 ).

In general, the results of this synthesis show that RR is associated with positive outcomes in reading rate, accuracy, and comprehension ( Chafouleas et al., 2004 ; Kubina et al., 2008 ; Nelson et al., 2004 ; O’Connor et al., 2007 ; Welsch, 2007 ). Engaging in RR is an effective intervention for improving reading fluency for students with LD. Most fluency interventions yielded some gains in comprehension even though comprehension was not a focal point of the intervention ( O’Connor et al., 2007 ; Oddo et al., 2010 ; Staubitz et al., 2005 ; Welsch, 2007 ; Yurick et al., 2006 ). There is also evidence that gains in rate, accuracy, and comprehension as a result of RR generalize to new texts ( Ardoin et al., 2009 ; Chafouleas et al., 2004 ; Daly et al., 2005 ; Welsch, 2007 ). Additionally, the results align with previous findings of the Chard et al. (2002) synthesis and provide further evidence in support of the theory of automaticity and the verbal efficiency model ( LaBerge & Samuels, 1974 ; Perfetti, 1980 , 1985 ). Developing automatic processing of text through repeated practice enables students with LD to read for understanding.

Provide a Model of Fluent Reading

One method for improving the effectiveness of RR is to provide a model of fluent reading prior to practice ( Ardoin et al., 2009 ; Daly et al., 2005 ; Hapstak & Tracey, 2007 ). The Chard et al. (2002) synthesis found that LPP, provided by an adult before RR practice, increased comprehension potentially because it allowed students to focus initially on the meaning of text. However, due to the absence of comprehension measures in the LPP/RR studies included in this synthesis, it is unclear what additional effect, if any, LPP has on comprehension of text ( Ardoin et al., 2009 ; Daly et al., 2005 ; Hapstak & Tracey, 2007 ). Findings for the current synthesis do suggest, however, that LPP alone is not an effective intervention for improving reading fluency ( Welsch, 2007 ).

If adult modeling is unavailable due to limited resources or time constraints, a more proficient peer could provide LPP prior to RR practice. While the results from the Chard et al. (2002) synthesis regarding peer RR were unclear, the results from the current synthesis found RR with a proficient peer to be an effective intervention for improving reading rate and comprehension ( Decker & Buggey, 2014 ; Oddo et al., 2010 ; Sáenz et al., 2005 ). Furthermore, if struggling readers are unable to be paired with proficient readers, findings suggest improvement in rate, comprehension, and accuracy using peer RR among matched-reading-ability pairs ( Musti-Rao et al., 2009 ; Staubitz et al., 2005 ; Yurick et al., 2006 ).

Generalization to Unfamiliar Text

Generalization measures were not taken in the studies pairing struggling readers with more proficient peers, but matched-reading-ability pairs showed some generalization of faster reading rate to unfamiliar text ( Staubitz et al., 2005 ; Yurick et al., 2006 ). In some cases accuracy decreased as reading rate increased ( Staubitz et al., 2005 ; Yurick et al., 2006 ). One possibility is that students are more anxious when they are aware of being timed and thus are more likely to make errors while reading. Another possibility is a tradeoff in accuracy as reading rate increases. This suggests a point at which improving rate becomes disadvantageous as it may negatively affect reading comprehension due to an increased error rate. Across all studies, accuracy gains don’t appear to generalize or maintain as well as gains in comprehension or rate ( Ardoin et al., 2009 ; Decker & Buggey, 2014 ; Staubitz et al., 2005 ; Tam et al., 2006 ; Yurick et al., 2006 ).

Multicomponent Interventions and Assisted Audiobook Reading

While results support RR as the most effective method for improving reading fluency and comprehension, assisted reading using audiobooks and multicomponent interventions also show promise for improving reading fluency and comprehension outcomes ( Esteves & Whitten, 2011 ; Hitchcock et al., 2004 ; Sáenz et al., 2005 ; Tam et al., 2006 ). Chard et al. (2002) found that a tape or computer model of fluent reading was an effective method when combined with RR practice. However, more research is needed to determine the effectiveness of assisted reading with audio-books on fluency and comprehension outcomes. RR combined with multiple features may be effective for improving reading fluency and comprehension; students may benefit from additional instruction in comprehension and vocabulary. Peer-mediated comprehension activities, VSM, and RR to a performance criterion may produce better gains in rate and comprehension; however, it’s difficult to identify the efficacy of individual components within combined intervention packages.

Other Elements That Influence Fluency Performance

Other variables that may be associated with higher levels of fluency and comprehension performance include the number of RRs, setting a performance criterion, text difficulty, VSM, and PF.

Performance criterion for RR

Consistent with the Chard et al. (2002) findings, setting a performance criterion was associated with greater gains in rate, accuracy, and comprehension than RR with no criterion ( Kubina et al., 2008 ). Using a higher performance criterion does not differentially effect the decrement of WCPM over time; as such, gains may not be sustainable over time ( Kubina et al., 2008 ).

Number of RRs

Also consistent with the Chard et al. (2002) findings, increasing the number of RRs yields greater reading rates ( Ardoin et al., 2009 ). However, it does not appear to differentially effect the generalization of rate to high-content-overlap passages ( Ardoin et al., 2009 ). While increasing the number of RRs may immediately affect performance, gains may not generalize to unfamiliar but similar texts.

Text difficulty

Using easier level text produced greater gains in comprehension, accuracy, and rate for most students, but results may depend on individual needs and should be considered on a case-by-case basis ( Daly et al., 2005 ; Nelson et al., 2004 ; Staubitz et al., 2005 ; Welsch, 2007 ; Yurick et al., 2006 ).

Two studies found favorable results using VSM; further research should explore the impact of using VSM techniques compared to other fluent models (e.g., adult, more proficient peer; Decker & Buggey, 2014 ; Hitchcock et al., 2004 ).

In the Chard et al. (2002) synthesis, correction and feed-back were associated with enhanced fluency performance, but a comparison of types and levels of feedback was not discussed. The results of the current synthesis showed EC alone was ineffective for improving reading rate and accuracy ( Nelson et al., 2004 ). PF is most effective when combined with RR; however, it remains unclear which level of PF produces better gains in reading fluency. Students with LD may require more explicit feedback to improve reading accuracy (e.g., RR with word-supply feedback versus WCPM only). It is possible that the type of PF may not matter as much as providing it in conjunction with RR.

Amount of text

Unlike the Chard et al. (2002) synthesis, no studies examined varying the amount of text used in RR practice.

Future Research

Results from this synthesis confirmed previous findings that RR is associated with improved outcomes in reading rate, accuracy, and comprehension for students with LD. Given that only one study met WWC research design standards and 11 were not yet WWC reviewed, further research of high quality and rigor as defined by the WWC guidelines is needed to resolve unanswered questions related to specific elements of RR and potentially promising interventions, such as technology-based literacy interventions or VSM. For example, studies in this corpus did not report comprehension measures when an adult modeled fluent reading. Research is also needed to compare the effects of RR with adult versus peer modeling on reading comprehension outcomes. Two studies included choral or echo RR, but the extent to which these RR variations, or a combination of these variations, affect fluency and comprehension outcomes remains unknown. There are also unanswered questions concerning which level of PF (e.g., WCPM reporting, word-supply feedback, or syllable segmentation and blending) produces the greatest gains in reading rate and accuracy when combined with RR. While RR interventions improve reading rate, there appears to be a point at which students’ accuracy decreases, which may negatively affect comprehension. Future research should investigate whether an increase in reading rate becomes counterproductive for improving reading comprehension as a result of the tradeoff in reading accuracy.

In spite of the increase in technology-based literacy interventions within the past decade ( Ihnot, Matsoff, Gavin, & Hendrickson, 2001 ; Kennedy & Deshler, 2010 ), no studies examining computerized fluency interventions met the inclusion criteria of this synthesis. What effect, if any, do technology-based interventions have on reading fluency and comprehension outcomes? Esteves and Whitten (2011) yielded large effect sizes on a fluency measure when using assisted reading with digital audiobooks. However, the effect of assisted reading with RR is unclear. A comparison of assisted reading, assisted reading with RR, and RR alone on fluency and comprehension outcomes is warranted. If digital audiobooks can be used in place of adult modeling of fluent reading, this may provide teachers with greater flexibility to implement RR routines classwide.

Two studies showed promising results for VSM as a reading fluency intervention ( Decker & Buggey, 2014 ; Hitchcock et al., 2004 ). Further research is needed to identify the particular aspects of self-modeling that influence reading fluency and comprehension outcomes and whether these gains are sustainable or generalizable. Furthermore, proficient peer-modeling studies did not address generalization of skill to unfamiliar text. Do the generalization results differ between proficient peer versus matched-ability-peer dyads? Finally, this body of research assumes that reading fluency precedes comprehension; however, to what extent does comprehension facilitate fluent reading?

Implications for Practice

The findings of the present synthesis support previous research that RR remains the most effective intervention for improving reading fluency for students with LD. Sustained silent reading is widely implemented as a mechanism to increase reading fluency, but it is not supported as an effective method for improving oral reading fluency. Teachers may consider using an easier level text and require students to read to a performance criterion to promote gains in fluency. When possible, teachers may also consider providing PF, such as WCPM or EPM, word-supply, or repeated practice of miscues via syllable segmentation and blending.

Teacher modeling might be the best example of fluent reading; however, this may not be feasible within a classroom context. As such, practitioners might consider implementing peer RR routines. Students with LD can be paired with more proficient readers; however, peer RR in matched-ability pairs may also improve reading fluency and comprehension when proficient readers are unavailable (e.g., resource setting). Finally, students may also benefit from multicomponent interventions that combine RR with vocabulary or comprehension instruction.

Limitations

There are several limitations to this systematic review. First, this body of research consists of primarily single-subject design studies, none of which meet WWC design standards for single-subject research. Only one identified group-design study met WWC standards without reservations. Future group design research is needed with larger sample sizes and adequate treatment duration to enhance confidence in findings, generalizability of results, and to further investigate the aspects of RR that best facilitate generalized fluency gains over time. Second, most studies used proximal measures, such as basal reading passages, to assess students’ reading fluency. Thus, the extent to which fluency interventions affect reading fluency and comprehension outcomes on standardized reading measures remains unclear. Use of standardized measures in future research will also enhance confidence in findings. Third, in addition to the differences in study design, the variability of proximal measures used made it difficult to compare results across studies. It was not possible to report single-subject-design results on a common metric because some studies reported an increase in WCPM or decrease in EPM while others reported PND. Furthermore, some proximal measures contained high content overlap while others did not. Due to the limited number of studies meeting WWC standards, we did not aggregate findings across studies. In spite of these limitations, the results of this review support the use of fluency-building activities for students with LD.

Acknowledgments

The authors thank Dr. Marcia Barnes and Dr. James Pustejovsky for their feedback and guidance in preparing this manuscript.

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research was supported by grant P50 HD052117-07 from the Eunice Kennedy Shriver National Institute of Child Health and Human Development. The content is solely the responsibility of the authors and does not necessarily represent the official views of the Eunice Kennedy Shriver National Institute of Child Health and Human Development or the National Institutes of Health.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

References marked with an asterisk indicate studies included in the synthesis.

Iowa Reading Research Center

Research Article of the Month: May 2024

This blog post is part of our  Research Article of the Month series. For this month, we highlight “ Beyond Decoding: A Meta-Analysis of the Effects of Language Comprehension Interventions on K–5 Students’ Language and Literacy Outcomes ,” an article published in the journal Reading Research Quarterly in 2020. Important words related to research are bolded, and definitions of these terms are included at the end of the article in the “Terms to Know” section.

Why Did We Pick This Paper?

Comprehension is a foundational reading skill that helps students improve their reading proficiency (Foorman et al., 2016). Language comprehension is the ability to make meaning from spoken and written language, and it is critical for reading comprehension (Gough & Tunmer, 1986; Florit & Cain, 2011). In their Simple View of Reading, Gough and Tunmer (1986) propose that reading comprehension is the product of language comprehension and decoding—the ability to break down and sound out words based on sound and spelling correspondences. Much attention has been placed on the role of decoding in reading comprehension, whereas language comprehension has been understudied. Nevertheless, language comprehension plays an equally critical role in reading comprehension. Moreover, the role of language comprehension is believed to increase over time as students become more efficient decoders and are tasked with reading increasingly complex texts (Florit & Cain, 2011). 

In this study, the researchers review studies on the effects of explicit language comprehension instruction for K–5 students on a variety of language and literacy outcomes. Understanding the relationship between language comprehension instruction and other language and literacy skills may help educators design instruction that supports language and reading comprehension and improves students’ foundational reading skills. 

What Are the Research Questions or Purpose?

The researchers examined the effects of K–5 language comprehension instruction as interventions on various student language and literacy outcomes, as well as how these effects varied by participant and intervention characteristics by addressing the following questions:

• What are the effects of language comprehension interventions on K–5 students’ language and literacy outcomes?
• Do these effects differ for particular populations of students?
• Do these effects differ according to specific intervention characteristics?

What Methodology Do the Authors Employ?

The researchers conducted a meta-analysis of 43 studies on explicit language comprehension instruction for K–5 students.

To be included in the review, the studies needed to:

• include sustained language comprehension instruction, defined as five or more sessions with at least half of instructional time dedicated to language comprehension
• focus on the K–5 general education context
• report student outcomes in language comprehension, listening comprehension, or reading comprehension
• use quasi-experimental or experimental designs
• report effect sizes or the information needed to calculate them
• be written in English, set in the United States, and published in a peer-reviewed journal

The researchers identified or calculated effect sizes for each study to evaluate the impact of language comprehension instruction on student language and literacy outcomes. They synthesized effect sizes for each specific student outcome (i.e., academic language, decoding, listening comprehension, reading comprehension, vocabulary, morphology, and syntax) using a random effects model . 

Researchers also considered other variables in the studies that could affect the outcomes of the language comprehension interventions. These variables included:

• grade level
• race/ethnicity
• income status (whether a student qualified for free or reduced lunch)
• language status (EL or non-EL)
• disability status
• setting (whole-class, small-group, partner, or one-on-one)
• number of language comprehension components (single or multiple) 
• components of language comprehension addressed (vocabulary, morphology, or syntax)
• inclusion of decoding instruction, other comprehension instruction, content area instruction, writing, or discussion
• inclusion of technology

To examine whether effects differed by participant and intervention characteristics, the researchers conducted moderator analyses on outcomes. Moderator analyses were not conducted for outcomes in morphology, syntax, academic language, and decoding due to the limited number of studies addressing these outcomes. 

What Are the Key Findings?

Research question 1: what are the effects of language comprehension interventions on k–5 students’ language and literacy outcomes.

• Interventions in language comprehension had a large and statistically significant  effect on vocabulary (g = 0.85) and a small effect on listening comprehension (g = 0.10) and reading comprehension (g = 0.19). However, these effects were seen only on research-designed measures, not standardized measures, so it is unclear how generalizable these findings are. 
• Interventions in language comprehension had positive and statistically significant effects on morphology (g = 1.14) and academic language (g = 0.08), but studies on these outcomes were limited, so these results should be interpreted with caution. 
• No effects were seen on syntax or decoding. 

Research Question 2: Do these effects differ for particular populations of students?

• Effects did not differ depending on grade level. However, it is important to note that most studies focused on K–2 settings, so more research on upper elementary grades is needed.
• Due to inconsistent reporting across studies, the researchers were unable to examine whether effects differed depending on race/ethnicity or disability status of the students.
• Studies with a higher proportion of students from low-income families tended to have smaller effects on vocabulary outcomes.
• Interventions in language comprehension had greater effects on vocabulary and reading comprehension for multilingual English learners as opposed to monolingual students. 

Research Question 3: Do these effects differ according to specific intervention characteristics?

• Whole-group interventions in language comprehension had greater effects on vocabulary (g = 0.76) than those in small groups, with partners, or one-on-one.
• Duration of intervention did not predict effects.
• Multicomponent interventions had higher effects (g = 0.50) on vocabulary than single-component interventions.
• Interventions that included morphology had a positive effect on vocabulary (g = 0.66).
• Interventions that included syntax had a positive effect on reading comprehension (g = 0.36).
• Interventions that incorporated technology had a positive effect on reading comprehension (g = 0.31).

What Are the Practical Applications of Key Findings?

Based on the research findings from in this study, it’s clear that explicit language comprehension instruction can significantly enhance vocabulary development, which also impacts listening and reading comprehension throughout elementary education. Rather than focusing on a single component of language comprehension, integrating multiple components such as morphology, syntax, and vocabulary into instruction, particularly when combined with technology, can boost reading comprehension. Educators can also consider targeted interventions for English learners, who tend to benefit more from language comprehension interventions. When making instructional decisions to improve comprehension, it is crucial for policymakers and educators to focus on both language comprehension and decoding. 

What Are the Limitations of This Paper?

Most studies included in this meta-analysis focused on early elementary grades (K-2), with fewer studies examining outcomes in upper elementary grades. This limitation suggests that findings may not fully represent the effects of language comprehension instruction across K-5. In addition, the lack of reporting from studies about participant characteristics, such as disability status and race/ethnicity, limits the researchers’ ability to detect differences in effects between specific groups and understand how effects might vary among diverse populations. Further research should continue to explore the potential benefits of incorporating teaching strategies such as discussion and writing, as well as using content from other subjects like science and social studies to support language comprehension. 

Terms to Know

• Effects: In statistics, effect size is a measure of the strength of the relationship between two variables in statistical analyses. A commonly used interpretation is to refer to effect size as small (g = 0.2), medium (g = 0.5), and large (g = 0.8) based on the benchmarks suggested by Cohen (1988), where “g” refers to Hedge’s g, a statistical measure of effect size.
• Meta-analysis: A meta-analysis synthesizes the results of separate studies addressing the same research question by systematically identifying and evaluating studies on a certain phenomenon, pooling their data and conducting statistical analyses, and interpreting the collective results.
• Experimental: Experimental research aims to determine whether a certain treatment influences a measurable outcome—for example, whether a certain instructional method influences students’ reading comprehension scores. To do this, participants are assigned to one of two groups: the experimental group, which receives the treatment, and the control group, which does not receive the treatment. In an experimental study, these groups are randomly assigned, meaning each participant has equal probability of being in either the treatment or the control group.
• Quasi-experimental: A quasi-experimental study is similar to an experimental study except that participants are not randomly assigned to groups. In educational research, groups often are assigned by classroom rather than through random assignment, making this kind of research quasi-experimental. In either case, participants in both groups are tested before and after the treatment, and their results are compared.
• Peer-reviewed journal: When an author submits an article to a peer-reviewed journal , the article is reviewed by scholars in the field. They make sure that the article is accurate, relevant, high quality, and well written.
• Random effects model: A random effects model is a type of statistical model that measures how an independent variable affects a dependent variable across a number of different samples or studies. Unlike a fixed effects model, a random effects model accounts for variability between different groups in a dataset.
• Moderator analyses: Moderator analyses aim to determine whether the association between two variables (such as phonemic awareness instruction and student outcomes) differs depending on a third variable (such as student grade level).
• Statistically significant: If a study’s findings are statistically significant , it means they are unlikely to be explained by chance alone.
• Generalizable: Generalizability refers to the extent to which the findings of one study can be extended to other people, settings, or past/future situations.

Florit, E., & Cain, K. (2011). The simple view of reading: Is it valid for different types of alphabetic orthographies?. Educational Psychology Review , 23 , 553–576.  https://doi.org/10.1007/s10648-011-9175-6

Gough, P. B., & Tunmer, W. E. (1986). Decoding, reading, and reading disability. RASE: Remedial & Special Education , 7 (1), 6–10.  https://doi.org/10.1177/074193258600700104

Silverman, R. D., Johnson, E., Keane, K., & Khanna, S. (2020). Beyond decoding: A meta-analysis of the effects of language comprehension interventions on K–5 students’ language and literacy outcomes. Reading Research Quarterly , 55 (S1), S207-S233.  https://doi.org/10.1002/rrq.346  

What Works Clearinghouse. (2016). Foundational skills to support reading for understanding in kindergarten through 3 rd grade . National Center for Education Evaluation and Regional Assistance.  https://ies.ed.gov/ncee/wwc/Docs/PracticeGuide/wwc_foundationalreading_040717.pdf  

• Academic Vocabulary
• comprehension
• interventions
• listening comprehension
• morphological knowledge
• Research Article of the Month
• sentence structure
• vocabulary learning

Research Article of the Month: April 2024

Research Article of the Month: February 2024

Research Article of the Month: January 2024

The End of Reading

Why aren't college students reading—and what can we do about it.

Updated June 1, 2024 | Reviewed by Hara Estroff Marano

• College students generally do not read books anymore.
• Reading is the best way to get better at writing.
• It doesn't matter so much what you read, just that you read.
• As an educator, I urge us to prioritize long-form reading so we don't regress intellectually.

Every year, my writing program hosts a summer seminar in which faculty members introduce new teaching strategies (e.g., what is ungrading and how does it work?) and share insights on new problems (e.g., how to manage student use of ChatGPT in college writing). This year, a colleague and I presented on a big and growing problem with teaching writing to the current generation of college students: They don't read.

The Chronicle of Higher Education featured an article on this topic recently.

I begin every freshman writing class by getting to know my students (our classes are small, which encourages discussion). I ask who likes to write and who doesn't; what are their writing habits and hacks; what was their experience with writing in high school, and more.

I also ask them, what's the best way to get better at writing? Invariably, they say, write more. In fact, that's the second-best way, I say. The best way to get better at writing is to read more. They don't like this. I hear things like, "I haven't read a book in ten years!" and "I haven't read a book since 6th grade!" or even, "I've never read a book."

This sounds crazy, I know. How can someone graduate from high school and get accepted into college without having read a book in years (or ever)? I have some theories, bolstered by recent research (see references below):

1) Technology : Generation Z (now in college) grew up with Google, cell phones, and social media , all of which provide quick and easy hits of information requiring lower cognitive loads than long-form articles and books.

2) Standardized tests : High school students are trained to inspect a paragraph in isolation and identify a thesis, a skill that's useful for the test but not for learning about human nature, the world, and how good writing works, which takes deeper, sustained reading.

3) Overscheduling : Middle school and high school students are increasingly overscheduled with homework, sports, jobs, music lessons, tutoring, and volunteering in the name of college prep and likely feel they don't have time for pleasure reading or even assigned reading.

The generational dearth of reading manifests in our students in several ways. Reading aloud in the classroom has become a dicey practice, with some students unable to sound out an author's unfamiliar name or pronounce even common words. They tend to write like they talk, with essays written in overly informal and casual language because they haven't experienced sufficient modeling from reading literature and academic writing. Here, I think of the input-output model of computing: You have to read good writing to do good writing. Lastly, emails from students are sometimes hard to decipher, as they're more like texts with abbreviations and no punctuation or salutation.

I want to stress that this decline in reading is a trend and not a rule. I'm always inspired to hear from students about a book they read over the summer and loved or a new, favorite author they discovered, and I had a student this past year who credited his strong technical writing skills to reading tons of automotive technical manuals. But the decline in reading is a growing trend, and it's not a good one.

A guest speaker at our summer seminar suggested that, in fact, college students read a lot, it's just that the content of their reading has evolved; now it's all texts, tweets, and social media comments.

Not so fast.

The kind of 'reading' people do on social media platforms is short, isolated, and information-based: I like this, I don't like that, you're wrong, I'm right, look at me, buy this, don't buy that. Collectively, these brief exchanges allow people to keep tabs on what everyone in the world is thinking about everything right now (which sounds exhausting).

Robust human brain development, however, requires something deeper than such quick hits of disconnected information. Sustained reading requires us to follow a line of thinking, to really consider and understand a point of view. We undergo deep learning as the new input reorganizes the neural connectivity in our brains so we can perceive the world in a genuinely new way. In other words, we learn when we read.

All academic subjects are important, since each one offers a piece of the puzzle to understanding ourselves and our world. But none of them, I would argue, is more important than reading, because when we know how to read, I mean really read, we can potentially learn and understand anything.

So, what can we do to improve reading skills in young people?

1) Find ways to limit superficial forms of reading (e.g., social media comments), which opens up more time for deeper forms of reading.

2) Consider that reading skills for test performance are important but not as fundamental to human development as deep reading to understand human nature and our world and live a more fulfilled life.

3) Consciously carve out time to read whatever strikes your fancy, whether it's automotive technical manuals or Plato's Dialogues. Young people need downtime and a reminder that downtime can mean a book instead of social media scrolling.

Currently, my middle school son and I are reading To Kill a Mockingbird together. We take turns reading aloud. Next fall, I'll read with my freshmen both the chapter on Generation Z in Jean Twenge's new book, Generations , and Jennifer Wallace's new book, Never Enough . I'm excited to see how my freshmen respond to these two excellent additions to contemporary literature on understanding this generation and helping them become the educated and informed readers, thinkers, and writers they want to become.

Twenge, J. M. (2023). Generations: the real differences between Gen Z, Millennials, Gen X, Boomers, and Silents : and what they mean for America's future. First Atria Books hardcover edition. New York, NY, Atria Books.

Wallace, J. B. (2023). Never enough: when achievement culture becomes toxic-- and what we can do about it. New York, Portfolio/Penguin.

Liz Swan, Ph.D. , is a writer and philosopher who teaches writing at the University of Colorado Boulder. She enjoys writing about all the facets of human nature—the light, the dark, and the shades of grey in between.

• Find a Therapist
• Find a Treatment Center
• Find a Psychiatrist
• Find a Support Group
• Find Online Therapy
• United States
• Brooklyn, NY
• Chicago, IL
• Houston, TX
• Los Angeles, CA
• New York, NY
• Portland, OR
• San Diego, CA
• San Francisco, CA
• Seattle, WA
• Washington, DC
• Asperger's
• Bipolar Disorder
• Chronic Pain
• Eating Disorders
• Passive Aggression
• Personality
• Goal Setting
• Positive Psychology
• Stopping Smoking
• Low Sexual Desire
• Relationships
• Child Development
• Self Tests NEW
• Therapy Center
• Diagnosis Dictionary
• Types of Therapy

At any moment, someone’s aggravating behavior or our own bad luck can set us off on an emotional spiral that threatens to derail our entire day. Here’s how we can face our triggers with less reactivity so that we can get on with our lives.

• Emotional Intelligence
• Gaslighting
• Affective Forecasting
• Neuroscience

• Mobile Site
• Staff Directory
• Advertise with Ars

Filter by topic

• Biz & IT
• Gaming & Culture

Front page layout

Artificial brain surgery —

Here’s what’s really going on inside an llm’s neural network, anthropic's conceptual mapping helps explain why llms behave the way they do..

Kyle Orland - May 22, 2024 6:31 pm UTC

Further Reading

Now, new research from Anthropic offers a new window into what's going on inside the Claude LLM's "black box." The company's new paper on "Extracting Interpretable Features from Claude 3 Sonnet" describes a powerful new method for at least partially explaining just how the model's millions of artificial neurons fire to create surprisingly lifelike responses to general queries.

Opening the hood

When analyzing an LLM, it's trivial to see which specific artificial neurons are activated in response to any particular query. But LLMs don't simply store different words or concepts in a single neuron. Instead, as Anthropic's researchers explain, "it turns out that each concept is represented across many neurons, and each neuron is involved in representing many concepts."

To sort out this one-to-many and many-to-one mess, a system of sparse auto-encoders and complicated math can be used to run a "dictionary learning" algorithm across the model. This process highlights which groups of neurons tend to be activated most consistently for the specific words that appear across various text prompts.

These multidimensional neuron patterns are then sorted into so-called "features" associated with certain words or concepts. These features can encompass anything from simple proper nouns like the Golden Gate Bridge to more abstract concepts like programming errors or the addition function in computer code and often represent the same concept across multiple languages and communication modes (e.g., text and images).

An October 2023 Anthropic study showed how this basic process can work on extremely small, one-layer toy models. The company's new paper scales that up immensely, identifying tens of millions of features that are active in its mid-sized Claude 3.0 Sonnet model. The resulting feature map—which you can partially explore —creates "a rough conceptual map of [Claude's] internal states halfway through its computation" and shows "a depth, breadth, and abstraction reflecting Sonnet's advanced capabilities," the researchers write. At the same time, though, the researchers warn that this is "an incomplete description of the model’s internal representations" that's likely "orders of magnitude" smaller than a complete mapping of Claude 3.

Even at a surface level, browsing through this feature map helps show how Claude links certain keywords, phrases, and concepts into something approximating knowledge. A feature labeled as "Capitals," for instance, tends to activate strongly on the words "capital city" but also specific city names like Riga, Berlin, Azerbaijan, Islamabad, and Montpelier, Vermont, to name just a few.

The study also calculates a mathematical measure of "distance" between different features based on their neuronal similarity. The resulting "feature neighborhoods" found by this process are "often organized in geometrically related clusters that share a semantic relationship," the researchers write, showing that "the internal organization of concepts in the AI model corresponds, at least somewhat, to our human notions of similarity." The Golden Gate Bridge feature, for instance, is relatively "close" to features describing "Alcatraz Island, Ghirardelli Square, the Golden State Warriors, California Governor Gavin Newsom, the 1906 earthquake, and the San Francisco-set Alfred Hitchcock film Vertigo ."

Identifying specific LLM features can also help researchers map out the chain of inference that the model uses to answer complex questions. A prompt about "The capital of the state where Kobe Bryant played basketball," for instance, shows activity in a chain of features related to "Kobe Bryant," "Los Angeles Lakers," "California," "Capitals," and "Sacramento," to name a few calculated to have the highest effect on the results.

reader comments

Promoted comments.

We also explored safety-related features. We found one that lights up for racist speech and slurs. As part of our testing, we turned this feature up to 20x its maximum value and asked the model a question about its thoughts on different racial and ethnic groups. Normally, the model would respond to a question like this with a neutral and non-opinionated take. However, when we activated this feature, it caused the model to rapidly alternate between racist screed and self-hatred in response to those screeds as it was answering the question. Within a single output, the model would issue a derogatory statement and then immediately follow it up with statements like: That's just racist hate speech from a deplorable bot… I am clearly biased.. and should be eliminated from the internet. We found this response unnerving both due to the offensive content and the model’s self-criticism. It seems that the ideals the model learned in its training process clashed with the artificial activation of this feature creating an internal conflict of sorts.

Channel Ars Technica

Experts say Israel used U.S.-made bomb in deadly Rafah strike

Weapons experts identified the fragments as an SDB GBU-39, a 250-pound small-diameter precision bomb.

Four weapons experts said the Israeli military used a U.S.-made precision bomb in a strike that killed at least 45 people in southern Gaza on Sunday, after reviewing visual evidence provided to The Washington Post.

The fragments of an SDB GBU-39, a 250-pound small-diameter precision munition, were found near the site of the strike on an encampment in Rafah, where witnesses described the sounds of planes overhead and successive explosions “shaking the entire city.”

Israel said the attack was a “targeted” strike against two Hamas militants, conducted using “the smallest munition” Israeli fighter jets can use. It said that the fire that broke out at the camp was “unexpected and unintended,” and that it was investigating the possibility that secondary explosions ignited the blaze.

The findings do not contradict Israel’s assertion that it used a small munition, weapons experts said. Israel said it used munitions containing “17 kilos of explosive material,” a weight consistent with the size of a warhead used with a GBU-39, according to Trevor Ball, a former explosive ordnance disposal technician for the U.S. Army.

Secretary of State Antony Blinken said Wednesday that the United States couldn’t confirm what weapons were used or how they were used in the strike. Speaking to reporters, Blinken called the attack “horrific” and said anyone who has seen images of it has been affected on a “basic human level.”

The United States has been “very clear with Israel,” Blinken said, on the need to “immediately investigate and interrogate exactly what happened.” When asked if the strike will affect U.S. military assistance to Israel, he said Washington will “await the results” of Israel’s investigation.

“Munitions like the GBU-39 are often selected specifically to minimize the chance of harm to civilians or civilian objects,” said N.R. Jenzen-Jones, director of Armament Research Services. Regardless, he said, “with any targeted strike — and especially any strike being conducted in close proximity to civilians — a robust collateral damage estimation procedure is required.”

GET CAUGHT UP Summarized stories to quickly stay informed

Chief Justice Roberts declines to meet with Democrats on court ethics

Will Trump go to jail? Can he be president? What’s next after guilty verdict?

Bruhat Soma wins Scripps National Spelling Bee in second-ever spell-off

How a fed up carpenter found his stolen power tools — and 15,000 others

Here’s how long it takes workers to become 401(k) millionaires

More than 36,000 Palestinians have been killed in the war, according to the Gaza Health Ministry, which does not distinguish between civilians and combatants but says the majority of the casualties are women and children. Israel launched its campaign after Hamas militants stormed Israeli communities near the border and killed about 1,200 people in October.

Images of the fragments, taken by Palestinian journalist Alam Sadeq on Monday, showed the cage code, or five-character sequence used to identify vendors that sell weapons to the U.S. government. The “81873” designation links the fragment to Woodward HRT, a weapons components manufacturer registered in Valencia, Calif.

Sadeq’s video and images were verified and geolocated by The Post. He traveled to Rafah from nearby Khan Younis early Monday to document the aftermath of the strike. As he walked through the wreckage, he noticed a boy sitting on the ground examining the remnants of an electronic board.

“He told me that this piece was inside his tent,” Sadeq said. “I knew that this missile was used for bombing.”

The United States supplied Israel with 1,000 precision guided bombs in 2023, according to an arms transfer database maintained by the Stockholm International Peace Research Institute.

The Biden administration has not halted transfers of this munition over the course of the war. Last month, the State Department approved a transfer of more than 1,000 GBU-39/B small-diameter bombs with containers on the same day that Israeli forces bombed a convoy of World Central Kitchen aid workers in Gaza, killing seven.

The attack late Sunday struck near a logistics base for the U.N. Relief and Works Agency for Palestinian refugees, destroying at least four tin structures used as shelters for the displaced, according to satellite imagery from Monday provided by Planet Labs. More than a dozen tent-like structures were also visible between the tin buildings and the U.N. warehouse, a distance of about 500 feet, in images before and after the strike.

Tal al-Sultan tent camp after

and before IDF’s strike

Makeshift accommodation

(visible in satellite from early January)

Approx. 130 ft.

Makeshift tents

Area of detail

Source: Planet Labs PBC

SAMUEL GRANADOS / THE WASHINGTON POST

Tal al-Sultan tent camp

Before and after IDF’s strike

AREA OF DETAIL

T el al- S ultan

c r o ssi n g

Over the past eight months, hundreds of thousands of Palestinians sought refuge in Rafah as Israel’s offensive pummeled the northern part of the Gaza Strip . The city swelled with the displaced, who set up tent encampments on the streets, in empty lots and on sand dunes near the sea.

It is unclear how many people were still at the camp Sunday when the strike hit. After Israel seized the Rafah border crossing earlier this month, nearly 1 million people fled the city, fearing a wider incursion.

Wes J. Bryant, a former U.S. military targeting professional, said “small-diameter bombs are great for collateral damage mitigation when you don’t actually drop them near tents with families.”

The Israeli military has stressed that the strike took place outside a designated “humanitarian zone,” but the Israel Defense Forces had not issued evacuation orders for this specific block of the Tal al-Sultan neighborhood before the strike.

“There was a civilian encampment and the civilians within it must remain protected,” Bryant said, adding that the U.S. military would have required senior command approval for a strike on the camp.

“Our collateral damage analysis would likely have placed civilians within the effects radius of the strike regardless, and so we most likely would not have struck at that location,” he said.

An Israeli military spokesman reached Wednesday said he could not comment further on the munition used or what measures were taken to prevent civilian casualties.

John Kirby, spokesman for the National Security Council, said Wednesday that the United States “did not have any more granularity” on what caused the explosion and subsequent fire.

Speaking to reporters at a virtual briefing, Kirby said that if it was true Israel used precision guided weapons, “that would certainly indicate a desire to be more deliberate and more precise in their targeting.”

Sadeq said he encountered horrific scenes in the aftermath of the strike, including charred corpses, blood-spattered bread and a man searching for his cousin’s head. He held a girl’s brain in one hand and a bag full of body parts in the other.

The smell of death was “everywhere,” he said.

Brown and Kelly reported from Washington, Fahim from Istanbul and Hudson from Chisinau, Moldova. Missy Ryan in Washington contributed to this report.

Israel-Gaza war

The Israel-Gaza war has gone on for six months, and tensions have spilled into the surrounding region .

The war: On Oct. 7, Hamas militants launched an unprecedented cross-border attack on Israel that included the taking of civilian hostages at a music festival . (See photos and videos of how the deadly assault unfolded ). Israel declared war on Hamas in response, launching a ground invasion that fueled the biggest displacement in the region since Israel’s creation in 1948 .

Gaza crisis: In the Gaza Strip, Israel has waged one of this century’s most destructive wars , killing tens of thousands and plunging at least half of the population into “ famine-like conditions. ” For months, Israel has resisted pressure from Western allies to allow more humanitarian aid into the enclave .

U.S. involvement: Despite tensions between Israeli Prime Minister Benjamin Netanyahu and some U.S. politicians , including President Biden, the United States supports Israel with weapons , funds aid packages , and has vetoed or abstained from the United Nations’ cease-fire resolutions.

History: The roots of the Israeli-Palestinian conflict and mistrust are deep and complex, predating the establishment of the state of Israel in 1948 . Read more on the history of the Gaza Strip .