Can scientists fill the science journalism void? Online public engagement with science stories authored by scientists

In many countries the public's main source of information about science and technology is the mass media. Unfortunately, in recent years traditional journalism has experienced a collapse, and science journalism has been a major casualty. One potential remedy is to encourage scientists to write for news media about science. On these general news platforms, scientists' stories would have to compete for attention with other news stories on hard (e.g. politics) and entertaining (e.g. celebrity news) topics written by professional writers. Do they stand a chance? This study aimed to quantitatively characterize audience interactions as an indicator of interest in science news stories authored by early career scientists (henceforth ‘scientists’) trained to function as science reporters, as compared to news items written by reporters and published in the same news outlets. To measure users' behavior, we collected data on the number of clicks, likes, comments and average time spent on page. The sample was composed of 150 science items written by 50 scientists trained to contribute popular science stories in the Davidson Institute of Science Education reporters’ program and published on two major Israeli news websites—Mako and Ynet between July 2015 to January 2018. Each science item was paired with another item written by the website’s organic reporter, and published on the same channel as the science story (e.g., tourism, health) and the same close time. Overall significant differences were not found in the public's engagement with the different items. Although, on one website there was a significant difference on two out of four engagement types, the second website did not have any difference, e.g., people did not click, like or comment more on items written by organic reporters than on the stories written by scientists. This creates an optimistic starting point for filling the science news void by scientists as science reporters.

trained to function as science reporters were compared to audience reactions to news items written by reporters (organic items) and published in the same news outlets.
The digital media and Web 2.0 allow access to accurate and updated data on consumers' engagement with content [47,48], and sometimes even influence editorial decisions on topics and item placement [49][50][51]. Previous studies have pointed to the disparity between what journalists think interests the public and readers' choices, mainly as regards the emphasis on public affairs issues [52]. Studies of public engagement with science content online have mainly focused on views (clicks) and comments. They have analyzed engagement in terms of different forms of interactivity offered by the online platforms (e.g. clicking, commenting, emailing, etc.) that varies between topics and is time and context sensitive [47,[53][54][55][56][57].
Although engagement data are used routinely in online newsroom decision making, little is known about the characteristics of public interactions with online science items [58]. This is due primarily to the difficulty of obtaining data (e.g., number of clicks and average time on page) which are kept confidential for commercial reasons.

Research field and context
The Davidson Institute,-the Educational Arm of the Weizmann Institute of Science in Israel has trained and employed graduate students, postdocs and research fellows in the sciences as writers for its website since 2006. In early 2014, an academic conference panel about science and risk communication in the online Israeli media 1 (The 6 th Israeli Science Communication Conference, (2015) Davidson Institute of Science Education, Weizmann Institute of Science, Rehovot (June 24-25)) hosted the editor in chief of the Mako 2 (www.mako.co.il) news website. As a result of that meeting the Davidson Institute began collaborating with Mako by publishing science items written by scientists involved on its website [59]. Mako is the third most visited Israeli News site (23.2M entries in the last quarter of 2016, [60]), which is owned and operated by 'Keshet', Israel's largest TV commercial broadcasting company. It offers news content alongside streaming of TV shows. This type of collaboration was then also extended to Ynet3 (www.ynet.co.il), Israel's most widely read news website (52.5M entries in the last quarter of 2016, [60]). Ynet is operated by the 'Yediot Aharonot' media group that publishes a daily tabloid newspaper in addition to the website and caters mostly to young audiences (aged [18][19][20][21][22][23][24][25][26][27][28][29][30][31][32][33][34] surfing on mobile devices. Both news sites provide freely available news content and have a landing page showing the latest news headlines, with links to sections on specialized topics (channels, also known as 'Tabs'), similar to sections in print newspapers. Channels include 'World news'; 'Local news'; 'Parenting'; 'Vacation' etc. Both news sites have a social media presence. Ynet has about 1.1M likes and 1.07 followers on its Facebook page; Mako has about 823 thousand likes and 800 thousand followers on its Facebook page. According to the sites' editors, content is published on the sites' main page and on the channels' individual page based on newsworthiness. Readers do not have to follow a page or 'Like' it in order to interact with the item on the news site Facebook page.
At the time of the research, both news sites had an interface that allowed readers on the news site to easily share and like items to their Facebook, Twitter and Google+ accounts ( Fig  1). During the research period 'Ynet' removed the option to 'Like' an item from the news site and both news sites removed the option to share on Google+.
According to the StatCounter survey the most prevalent social media platform used in Israel (September 2019) is Facebook (67% of social media users in Israel) [61]. The news sites opte to give their readers indications of item popularity using only Facebook parameters in the form of counts of 'Likes' (Ynet) or 'Recommend' (Mako). The 'Like' (or 'recommend' on Mako) and sharing options are used differently. In order to mark an item with a 'Like', the reader clicks the 'thumbs up' icon, in order to share the item, one needs to click the platform's logo (when hovering above it the indication "share on. . ." appears), and it requires logging in to the user's account. The shared item will then appear on the person's wall. Hence, we only address the 'Like'/ Recommend parameter of the social media interface. The two websites do not employ a dedicated science journalist, or require the reporters covering these topics to have background in science.
The Davidson reporters' program currently employs about 50 graduate science students and faculty. The writers' program training does not have a defined syllabus or mandatory participation in a structured program. Rather, the scientists who participate in the program can sign up for a brief one-day training course twice a year, led by the editorial team. It teaches journalistic concepts and practices (e.g., avoiding jargon and passive voice) to familiarize these scientists with media style and formats and give them the tools to write their items successfully. Most of the training is done 'on-the-job' on an individual basis, between the scientists and the in-house editors as part of the editing process. The scientists contribute their critical reading and thinking, their proficiency and acquaintance with the research world and the ability to read original papers, understand them and their limitations. During their training, the scientists are taught about the journalistic creed in addition to practices that help them shift their attention [from an analysis of the science] to their item's impact on their readers [10]. However, it is important to emphasize that the scientists enrolled in the reporters' program are first and foremost graduate students and faculty members, they do not plan to pursue a journalistic career, but rather want to acquire additional soft skills. The Davidson editorial team is composed of science editors, two content editors and an editor in chief who is a former journalist. All the editors, except one content editor, have academic science degrees. The scientists have backgrounds in different fields and are at various stages of their graduate degrees, although a few are already faculty members. There is no quota demanded of each writer, but most scientists write between one to four 500-word items a month. The topics span the science, technology, mathematics and engineering (STEM) fields, and are chosen by the editorial team as a function of their newsworthiness and potential for public interest (or based on topics suggested by the scientists). The scientists are not allowed to write about their own research or research done in their labs, but are encouraged to write about local Israeli research as part of the Davidson agenda. The scientists are employed on an hourly basis to promote thorough inquiry (rather than being paid on the basis of number of words).
The Davidson Institute initially proposed this collaboration with the Mako and Ynet news site editors to increase the quantity and quality of science content in the news. According to Davidson staff, this collaboration enables scientists to share accurate, innovative scientific information and make it part of public's everyday news consumption while the news sites benefit from free high-quality science content. To date, this arrangement involves most of the mainstream news sites operating in Hebrew in Israel.
The editorial process: Each item submitted to the editorial panel is first sent for scientific editing that primarily involves fact checking and completing missing data. Next, the item is sent for content and linguistic editing. Finally, the item is given to the editor in chief for the final revision, who then approves it for publication. In cases where content editors have reservations, they consult the scientist who authored the item before sending it to the editor in chief. The editing process can last from a few hours to several days depending on the topic and format. The main job of the editorial panel is to provide accurate information, sometimes even at the expense of missing the deadline or a headline opportunity. The news site editors are provided with edited text, which they are not allowed to alter without Davidson's permission, but they are free to change the headlines as with any organic item. Although news editors have their own mandate on the headlines, there is an understanding between the news editors and the content providers that the approved headlines should not be misleading or commercialized.(For a critical analysis see also [59]). The name of the scientist appears in the credit line, and is visible to the readers even before clicking the item to read, and includes the person's title and affiliation to the Davidson Institute. The name of the writer, his or her affiliation and status is also stated at the end of the item (e.g. "Yael Groper, Davidson Institute of Science Education website reporter and a doctoral student in the Weizmann Institute of Science") alongside a link to the research article, if the item is based on one.
Researcher positioning: The first and third authors are university-based science communication researchers not affiliated with either Davidson, Mako or Ynet. The second author is the head of the science communication unit in the Davidson Institute and the initiator of the scientist writing program and collaboration with the news sites.

Data collection
As part of this research-practice collaboration, the researchers were given access to confidential Google Analytics data, including clicks, time spent on the page, likes and comments on Mako's and Ynet's websites. Due to the commercial sensitivity of the information, data mining took place on several consecutive days at the news company offices at Mako. The researcher was not allowed access to the information directly from Ynet but was sent the requested data electronically. Data were kept on an encrypted drive with access restricted to the first author alone. Due to the commercial nature of the data, the researchers were committed not to disclose the raw data. Hence, here, only descriptive data such as averages and standard deviation are presented, (see S1 Table for anonymized numeric data).
The initial dataset was composed of all 296 news items written by scientists and published on the two websites from July 2015 to January 2018 (Table 1). These were published mainly on the Digital, Health, Animals, Culture and Traveling channels (ranked by the number of items in each section).

Database of matching items
In order to compare public engagement, for each item written by scientist, an item written by an organic reporter e.g. the news site's reporter was paired in a matching process to yield the research database of a total of 150 pairs of news items.
Matching process: Each science item written by a scientist was paired with a corresponding news item written by a professional reporter that was published on the same channel and within an average of 0.8 days from it (about half of the items were published on the same day, and the rest no more than three days before or after); see Fig 1 for an example of paired items. After this initial filtering by channel and date of publication, we paired similar formats (e.g. quizzes, video articles, short/long items, etc.) when there was a choice of several organic items. For example, an item written by a scientist that reported on the passing of the last person who walked on the moon in the 'World news' channel was paired with the item "All the President's men. This is Trump's hive" published on the same day on the same channel (pair number 33 on 'Ynet', see S1 Table). In cases where no corresponding item was found on the same channel and within the designated timeframe, the reference item was excluded from the database (n = 57). Due to restricted access of the researcher to the Ynet news site's data and to broken links in both news sites, another 89 pairs were omitted from the sample. Overall, the process yielded a total of 150 pairs of news items, 51% of the initial dataset. The final database is a representative sample of the full database consisting of all channels and within the same time range as the full database.
On Mako website, 69% of items written by scientists were published on the Health channel, alongside site reporters' items on new treatments and diet suggestions (e.g. "Five delicious recipes for a healthy meal", 21/8/2017). Another 12% of the scientists' generated items were published on the Holiday/Travel channel and paired with items on new popular vacation resorts, celebrities' vacations and other travel information (e.g. "Where do the residents of the "Big Brother" reality show love to go on vacation?", 1/1/2016). An additional 7% were published on the HIX channel devoted to "Scientific discoveries, interesting phenomena, funny inventions, exciting news and other events from the world", and on the Culture channel.
On Ynet website, the vast majority of items (97%) were published on the Digital channel, alongside items such as "After a mouse and a gorilla: was a shark photographed on Mars?" (27/ 3/2016). The remaining 3% were published on the Animals, Health and World channels. While the scientists' items were always about a scientific study, or science related issues, the paired organic items were more diverse in terms of topics. Although the paired items had a similar topic, since they were published on the same channel, an organic item on the Digital channel on Ynet for example, could be a set of pictures showing readers what the Earth looks like at night, without any scientific or research related content. In four rare occasions both paired items covered the same exact topic side by side (one on Mako and 3 on Ynet, see Table 2 for item headlines). About half of the pairs in the database were both scientist and organic reporter items, on a scientific or research related topic. Since we could not pool the paired items from the two websites, there are too few items from each site to calculate statistical significance.

Data analysis and engagement types
Four quantitative parameters were chosen as indicators of audience engagement based on previous studies [43][44][45][51][52][53][54][55] and the available data. The data used in this study relied solely on absolute numbers for each parameter. Other relevant variables, such as an item's location on the website, length of time visible on the channel, length of the item etc., were not available. Clicks: number of clicks on the item's headline to view its content. Used in this study as a proxy for the number of readers. Indicates interest in reading the item based on its headline and a secondary sub-headline visible to readers before clicking (for average clicks per site see Table 3). Average Time on page: reported in seconds, from the opening of the item after clicking, until exiting the item page. A proxy for the time devoted by readers to each item. (for averages per site see Table 3). Likes: number of readers who clicked on the 'Like' button of the item. A measure of readers' favorable opinion of the item or the event it describes. It demands a higher engagement level on the part of the reader since by clicking 'Like' (in the Mako website, the 'Like' option is called 'Recommend', but these are equivalent), the item is published on the reader's wall on Facebook, thus exposing it publicly. According to "Facebook for developers" site, the 'Like' and 'Share' functions differ in terms of users' ability to add a personalized message and customize who they share the content with when using the 'share' button. The 'Like' and 'Recommend' options are the same in terms of function and it is the website's decision what version to display on their site. (for average 'Likes' per site see Table 3).
Comments: number of comments left by readers at the bottom of the item. Comments could be one word long to several paragraphs long, and could be off topic. Any internet user can post a comment anonymously on these two websites. (for averages of comments per site see Table 3).

Statistical analysis
Each engagement type was tested for normal distribution. A paired sample t-test was used for normally distributing parameters, such as average time on page on both websites, comments on Mako and Likes on Ynet. A Wilcoxon signed-rank test with a 95% confidence level was used for parameters that were not normally distributed, such as Clicks on both websites and Likes on Mako and Comments on Ynet (α = 0.05). In order to verify that the time difference Table 2. Headlines of paired items on the same topics.

Mako
"The HPV vaccination is perfectly safe" (21/10/ 2016) ""The vaccination doesn't cause paralysis that same day"" (20/10/2016) Four rare occasions in which an item on the same topic was published written by the website's organic reporter and also by a scientist from the reporters' program. Items addressed the same topics and were published more or less at the same time. These items differed in terms of the frames and angles the writers chose to take. Public engagement was higher with the items written by reporters on the first and last items, whereas the opposite was found for the two others. between the two sampling cycles did not affect the data, a one-way ANOVA was run on the first and last two months and the median date of the data collection cycles for each parameter. No significant difference was found. Therefore, the results of the statistical analyses are presented for the two rounds of data collection combined.

Results
In this study we are in the unique position where non-significant differences between groups are highly informative for the goals of the study. Fig 2 presents the comparison of all engagement types on both websites. In the case of the Mako website no significant differences were found between items written by scientists and Mako's organic reporters for average time on page and 'Likes', based on 92 and 85 pairs of items, respectively (Table 3). On the other hand, a Wilcoxon signed-ranks test showed a statistically significant difference in the number of Clicks (views) between items written by scientists and organic reporters (n = 89, Z = -2.862, p = 0.004) and a paired sample t-test showed a statistically significant difference in the number of Comments (n = 89, p<0.05) with more public engagement in response to Mako's' organic reporters on both parameters. To conclude Mako's results, there was no difference in the time devoted to reading the items or liking them (hitting the 'Like' button) but there was a clear preference to click items that were not written by scientists and were not necessarily about science.
While Mako's results were mixed, on the Ynet news site an analysis of the data retrieved on the paired items showed no significant differences between items written by scientists and organic reporters on any of the parameters (Table 3). For example, the average length of time on page for items written by organic reporters was 17:26 minutes, whereas the time on page for an item written by scientists had a viewing duration of 17:32 minutes, on average, as shown in Fig 2. Thus, based on our data, the public's interactions with science news written by scientists were not significantly different from other news items written by reporters on Ynet's news site.

Discussion
Americans' and Israelis' primary information source about science and technology is the online news medium [4,62], which is impacting public attitudes, perceptions and even behavior [1,[63][64][65][66][67]. News media has the potential to promote scientific understanding, especially if sufficient explanations of the science are provided and a narrative is used for its presentation [57]. Hence, accurate, accessible and relevant science stories in the news media are important factors in a healthy science communication environment. Unfortunately, given the current collapse of traditional journalism worldwide, science journalism has become a major casualty, thus hindering its ability to provide quality science coverage. For example, between 2013 and 2015, the number of science reporters in the Israeli news media decreased from 9 to only three, a drop that continues to this day [68]. The results of this study suggest that filling the void created by the firing of titular science reporters by scientists who write for the media may be a viable solution resulting in a higher frequency of scientific content in the news media which is also attractive to readers. This study examined whether readers reacted differently to science news items written by scientists as compared to news items written by organic reporters published on the same online news media sites. Generally speaking, based on our findings, the answer is no: audiences interacted similarly with both. This finding justifies the time and effort invested by the scientists and the Davidson science communication team to write attractive science stories, and justifies the resources provided by the news sites. Apparently if websites publish it, audiences will consume it. The website gains highly credible, reliable, science items free of charge and the scientists get a spot in a high exposure platform [69]. It could be claimed that our conclusion paints the results in excessively positive hues, since scientists' writing about science should be more accurate and knowledgeable than organic reporters, and therefore more popular among readers who enjoy its higher quality. Hence receiving a similar (rather than higher) level of reader engagement to that obtained by the sites' reporters is not a positive outcome for scientists who write for wider publics about science.
This view, however, makes several assumptions about the quality of scientists' writing (which we did not assess in this study), as well as about the key criteria of public enjoyment and appreciation of popular science writing on general interest news sites. Above all, it does not take the professionalism of journalists in their trade into account or the challenges of communicating science to a public that does not seek it out.
Our interpretation of the findings is therefore an optimistic one: although It is reasonable to assume that items written by scientists have higher density of science content, and are more accurate than the site's own reporting about science (neither employ a science writer), and although they compete for readers' attention with other topics, they nevertheless received a similar level of engagement from the public. Readers are willing to engage with scientific content authored by scientists, which we see as a very positive result.
These optimistic results raise a normative question about the impact of these collaborations on science journalism and science communication. In a democratic society the media do not only serve as an information conduit for the public [70], but also as platform to critique the authorities as regards misconduct, corruption or misuse of public resources [17,70].Scientists communicating science while working as scientists, and being part of an institution, may fill the informing and popularizing void, but find it very hard to take on a watchdog role. They cannot be reasonably expected to take the critical independent outsider approach of journalists, for example, by asking embarrassing or difficult questions (as would be expected from a journalist pursuing a lead) when interviewing a fellow scientist [21,36]. Providing ready-made scientific content to news outlets without charge may perpetuate the media's reliance on free, external content, and hence contribute to the weakening position of science journalists. It could be argued that accustoming the media to getting ready-made content without journalistic scrutiny may in fact be advancing 'churnalism'-a practice in which pre-packaged stories provided by news agencies and press releases are adapted for publication instead of reported news, and therefore potentially posing a danger to the legitimacy of science journalism and undermining its credibility [12,13,71].
While these are important caveats, it is crucial to note that although the Israeli online news media form one of the most accessible sources of information about science and technology related topics, they are also ranked as mostly unreliable (an average of 5.6 on a 10-point scale). University scientists, on the other hand, are seen as the most reliable source of information (an average of 7.6 on a 10-point scale), but inaccessible (only 19% of the survey respondents mentioned them as their sources of information about science and technology) [66]. The model described here integrates the strengths of each source; i.e., exposure and accessibility on online media and the reliability and expertise of university scientists. We present one model in which 'filling the void' can be achieved by early-career scientists who receive on-the-job training and editorial support that enables them to communicate science on general news sites. Needless to say, this type of partnership is not the only model. In recent years scientists have been communicating with diverse publics directly, thus avoiding the mediation of the news media altogether [2,24,25,72]. However, the model described here has already affected the availability and accessibility of science coverage in the general Israeli media.
The results from the two sites were mixed. No significant differences were found between scientists' and reporters' items on Ynet, but mixed findings were found on Mako, where reporters' items had more views and comments, but likes and time on page did not differ. The differences between the results from the two news sites may be attributed to their characteristics: Mako is owned by 'Keshet', Israel's largest commercial TV broadcasting company and shares the same website (www.mako.co.il) with television content on demand. Ynet is operated by the 'Yediot Aharonot' media group that also publishes a daily newspaper. This may suggest differences in the expectations and motivations of audiences reading the two websites. Although both websites offer news coverage on an array of topics, most visitors to the Mako website expect more entertaining options than when accessing Ynet for more hard news content. Given these differences in press orientation, our assumption is that more Mako readers accessed the website with light entertainment in mind, rather than an interest in finding out about recent scientific or technological developments. Another explanation is related to the structure of the websites. While Ynet has a designated channel for science in which the majority of the items in the database were published (97% of the items on Ynet, 39% of the total database), Mako does not publish this type of channel, and all the items from the database were published in the different channels the website offers its audience.
This study was enabled by a Research-Practice-Partnership (RPP). This unique position allowed the researchers access to data that are usually unavailable for commercial reasons, while providing practitioners with an evidence-based evaluation of their science communication efforts and the public's interaction with their products. Such RPPs have immense potential for improving practice and tapping experience-based questions and real-life data. More of these collaborations will increase reliance on behavioral data that can complement self-report research instruments.

Research limitations
One of the key limitations of this study was its reliance on the Google Analytics data mining system. The algorithm used to collect data by Google is unknown, and so are the basic assumptions underlying its data mining algorithm. Hence this study relied solely on absolute data for the number of clicks (views), Likes, etc. We did not use other information provided by Google Analytics such as age and gender since these constitute inferable data that rely on Google's search and deduction algorithm.
Our data show a significant difference in average time spent on page between the two websites. This could also be a result of the settings, preferences and specifications each news company used to configure the data collection. These were not disclosed either. This problem was mitigated by not comparing public engagement between the two websites, only between writers on the same website.
Another limitation relates to the editorial process. Specifically, we could not assess the effect of the editorial revision by the Davidson professionals on the attractiveness of the scientists' written output. However, this is also true of the items authored by the organic reporters whose work is also sent for editing by the professional news sites' editors that implements a similar process.
Finally, it is important to emphasize that this study focused on the quantity and type of audience interactions with two types of coverage. The quality of the coverage and user generated content was not addressed. For example, comments might only contain a title with no additional text, or be several paragraphs long. Comments could be positive or critical (e.g., 'interesting but it was hard to understand') or off topic. The quality of these aspects might defer between the types of coverage and their associated reader comments.

Conclusion
This study examined the public's interactions with scientists' popular writing, thus shedding light on online public engagement with science outside of an experimental setting. The results paint a positive picture where in most cases no differences were found between the ways audiences responded to scientific reports written by scientists-as-science-reporters, and stories written by news site reporters. One of this study's main conclusions has implications for Israeli news editors, who are reported to perceive science news as not being interesting to the public [3]. Specifically, this was the attitude of Mako website's editor [4]. The findings may contribute to changing this perception and result in the publication of more science related news in the general news, and should encourage scientists to write for the public on media outlets. This model may thus provide a practical solution for filling the science journalism void on struggling news media.
Supporting information S1 Table. Paired items datasets with 'Google analytics' data. Data of date of publication, number of clicks (views), 'Likes', Comments and average time on page is presented for each pair of items (e.g. scientist or reporter) on both news sites. (DOCX)