Google Reviews Explore Post-pandemic Shifts in Public Perceptions of Early Care and Education

---

Findings

This section presents detailed findings on public perceptions of care, both pre- and post-pandemic. We review the major implications of these findings in the “Discussion” section later in this brief. Throughout this section, we include sample direct quotes and excerpts from the Google reviews for each of the eight topics in this study. They are intended to illustrate the types of reviews included in this analysis.

Table 1 includes a list of eight identified topics of Google reviews related to child care, example keywords associated with each topic, and the number and percent of reviews that had these key terms in 2018/2019 and 2023.

---

Table 1. All topic areas except ‘cost’ increased in the prevalence of reviews post-pandemic.

Topics related to child care reviews, example keywords applied to reviews, and prevalence of topics

Source: Early, D. & Li, W. (2020). Google and Yelp reviews as a window into public perceptions of early care and education in Georgia. Child Trends.; Ekyalongo, Y. Y., Li, W., & Franchett, A. (2023). As the number of home-based child care providers declines sharply, parents are leaving more negative online child care reviews. Child Trends. https://doi.org/10.56417/962z2356o

---

The order of prevalence for each topic (i.e., which topics were mentioned the most, least, etc. in reviews) was the same in both 2018/2019 and 2023. The most prevalent topic at each evaluation period was warmth (mentioned in 77% of reviews in 2018/2019 and 85% in 2023). The least frequently discussed topic for both timepoints was cost (11% in 2018/2019 and 9% in 2023). All topics, except for cost, showed a significant increase (p<0.05) in mentions from 2018/2019 to 2023. For example, in 2018/2019, the topic of academics related to child care providers appeared in 42 percent of all Google reviews, compared to 51 percent of reviews in 2023. Cost was the only topic that decreased in frequency of mentions from 2018/2019 to 2023. The percentage of reviews that mention each topic at each timepoint appear in Figure 1.

---

Figure 1. Reviewers mentioned almost all topics significantly more post-pandemic than pre-pandemic, but mentions of cost were less frequent post-pandemic.

Percentage of reviews that mention each topic at each timepoint, 2018/2019 and 2023

Note: * indicates a statistically significant shift at p<0.05.

---

Next, we analyzed the percentage of reviews that were positive and negative at each timepoint to better understand the feeling or attitude about the program behind the words of the review. In total, we observed 19,901 positive (4 or 5 star) reviews and 4,608 negative (1, 2, or 3 star) reviews, or nearly five times more positive than negative reviews. Six topics showed significant increases (p<0.05) in the percentage of reviews that were positive (i.e., academics, convenience, safety, hygiene, communications, staffing) from pre- to post-pandemic. For instance, a significantly higher percentage of reviews about safety were positive in 2023 (80%) than in 2018/2019 (76%). Interestingly, cost was the only topic with a significant decrease (p=0.03) in positive reviews from 2018/2019 to 2023. (Differences in warmth were not statistically significant.) Figure 2 shows the percentage of reviews that were positive for each topic.

---

Figure 2. Most topics had a higher percentage of positive reviews post-pandemic than pre-pandemic, while cost had a significant decrease in positive reviews post-pandemic.

Percentages of reviews that were positive for each topic and timepoint

Note: * indicates a statistically significant shift at p<0.05. (All differences, aside from warmth, are significant.)

---

Finally, we examined the most frequent keywords used at each timepoint, both overall and for each topic. This analysis was done to examine whether there were differences in specific words being used at each timepoint; if so, we tried to understand what those changes might indicate in shifts of parents’ perceptions of care.  We counted keywords with different forms (e.g., “loving” and “love”) as part of the same keyword. Seven of eight topics had the same top keyword in 2023 as in 2018/2019, with the word “love” being the most frequent word at both timepoints (see Table 2). Only one topic, communication, had a different top keyword in 2023: The top communication keyword was “ask” in 2018/2019 and “update” in 2023. Notably, no keywords from the topic “safety” appear in the top 10 in 2018/2019, but “safe” appears in the top 10 for 2023. Table 2 presents the top 10 overall keywords at each timepoint.

---

Table 2. Top 10 overall keywords found in reviews at each timepoint

---

Limitations

While these findings offer a window into the public’s perceptions of child care, the data are not representative of parents’ attitudes about child care. What the public chose to post about drives the formation of topics used for this analysis, but this is often the result of a parent or caregiver’s strong reaction to a facility, classroom, or individual such as a teacher or director. Further, based on the star ratings, there were more positive reviews to analyze than negative reviews. Over 80 percent of reviews were 4 or 5 stars and thus categorized as positive reviews of ECE programs. This means there were fewer negative reviews (i.e., 1-, 2- or 3-star ratings) to analyze for change in sentiment over time compared to positive reviews. This could be because most parents having positive feelings about their child care arrangements, or it could be that respondents with more positive feelings about ECE programs appear more likely to have left a review than those with more negative feelings (which is common with online reviews). It is possible that fewer parents leaving average or negative reviews biased the topic modeling approach, drawing out more positive words and categories than truly exemplify parents’ thoughts on child care.

Lastly, we coded topics as being present in a review based on the presence of keywords and forms of keywords (e.g., “teach” for “teaching”), but this can lead to false positives. For example, “pay” is a reasonable keyword for the topic “cost,” but—as a result—reviews in which parents mention “paying attention” were coded as “cost,” even if the review did not mention the actual cost of child care. As presented in Early & Li (2020), words such as “attention” can refer to a teacher’s attention to a student, a parent’s attention to their child, or the attention to detail given to the facility. Additionally, the assignment of keywords to topics (and topic names themselves) were determined by the authors, not by the reviewers leaving online reviews. It is possible that the individuals leaving the reviews might differ in their opinions of how they would categorize their words in the reviews.

---

Discussion

The aim of this study was to provide a preliminary exploration into whether parent perceptions of access to child care shifted following the COVID-19 pandemic. Findings indicate that the order of prevalence for topics remained consistent between this analysis’ two time points (i.e., the order of topics in both 2018/2019 and 2023 was: warmth, staffing, convenience, academics, communication, safety, hygiene, and cost). This may indicate that, despite the massive impact of the COVID-19 pandemic on child care and the ECE workforce, parents remain fairly stable in their views about the importance of each topic area, which broadly correspond to the four family-centered dimensions of access (i.e., affordability [cost], supports child’s development [warmth, academics, safety, hygiene, staffing], meets parents’ needs [convenience, communication], and requires reasonable effort [also convenience, communication].

Both pre- and post-pandemic, the two most prevalent topics in Google reviews of care providers were warmth and staffing—highlighting the continued importance of a high-quality ECE workforce that can support children’s development, and that parents trust to care for their children. Within the topic of warmth, the most prevalent keyword was “love”; within the topic of staffing, the top keywords at both timepoints were “staff” and “teacher.” The two least prevalent topics in the Google reviews were cost and hygiene. In addition, the topic of cost was the only topic to be mentioned less frequently in 2023 than in 2018/2019.

Interestingly, hygiene-related reviews increased slightly from 2018/2019 to 2023, but did not increase as much as we expected following the COVID-19 pandemic. While the topic of hygiene was raised more often and more positively in reviews in 2023, it was still only mentioned in 15 percent of reviews—well below the prevalence of nearly all other topics. This might be because, by 2023, parents’ focus was shifting away from the pandemic. For this study, we did not analyze reviews posted during the COVID-19 pandemic (2020-2022), but it is possible that topics such as hygiene were more prevalent reviews from that era.

Cost was the only topic to have more negative than positive reviews in 2023, relative to 2018/2019, perhaps an indication of frustration for parents as inflation remains a concern and child care is more difficult to find. We examined the percentage of positive (i.e., a 4- or 5-star review) and negative reviews (i.e., a 1-, 2-, or 3-star review) at each timepoint to better understand the feeling or attitude behind the review’s words. All other topics had more positive than negative reviews in 2023, relative to 2018/2019, making the rise of more negative reviews about cost even more striking.

Safety, convenience, hygiene, and communication are topics that parents seem more focused on in 2023. All four topics increased from 2018/2019 to 2023 in both prevalence of mentions and in the proportion of reviews that were positive (i.e., attached to a 4- or 5-star review). Analysis of reviews indicates that parents have a tendency to post positive reviews when child care programs are meeting children’s basic needs such as safety and hygiene, and when they meet parents’ needs such as convenience and good communication.

In 2023, online reviews were more likely to raise the topic of safety than in 2018/2019, and the keyword “safe” became one of the top 10 keywords in 2023—highlighting parents’ increased attention to safety. In contrast, none of the keywords associated with the topic of safety rose to the top 10 list in 2018/2019. The top three keywords—“love,” “staff,” and “teacher”—were consistent across both timepoints.

Findings from this study provide a glimpse into parents’ values and priorities related to child care for their children. The methodology used for this research, topic modeling of online reviews of child care, can provide more information than just over-time analyses. Future research may consider analyzing how perceptions of child care differ by location or type of care, in addition to shifts in perceptions over time. As states continue to work on quality improvement efforts—including revisions to their Quality Rating and Improvement Systems (QRIS)—and improving consumer education, studies such as this could help them better understand parents’ needs and expectations for their children’s care and education.

---

Methodology and Data

This study builds off work from two previous studies—one that examined public reviews of ECE programs on Google and Yelp in Georgia, and another that scraped and reviewed Google, Yelp, and Care.com data nationally to analyze trends in negative and positive child care reviews. For this study, we extracted 24,509 publicly available Google ratings and reviews of 11,731 ECE programs across the United States. We defined ECE programs as programs that Google Maps listed as “child care,” “preschool,” or “day care.” Total reviews included 14,915 reviews from 2018-2019 (scraped in a previous study; Ekyalongo, Li, & Franchett, 2023) and 9,594 reviews from 2023.

We used a topic modeling approach to identify topics related to parents’ perceptions of ECE access pre- and post-pandemic from their reviews of child care programs. A topic modeling approach is a statistical technique for extracting topics from bodies of text, using the Natural Language Processing (NLP) technique to find keywords that commonly appear together, such as payment and money. A computer algorithm groups keywords together based on how frequently they appeared together, after which a research team labels the groups of words into a topic (e.g., the computer grouped the keywords payment and money, and the research team labeled this grouping as the topic of cost). For this study, we leveraged topics and keywords identified in a previous study of online child care reviews (Early & Li, 2020), rather than generating more topics (see Table 1). These topics were organized to align with the four dimensions of access: affordability (cost), supports child’s development (warmth, academics, safety, hygiene, staffing), meets parent’s needs, and requires reasonable effort (convenience, communication).

Recent research suggests that topic modeling may have limitations in processing short text excerpts, like social media data and online reviews. Therefore, to strengthen the results of topic modeling, we utilized generative AI models (GPT-3.5 Turbo) to facilitate two iterations of assigning topics. First, we applied topics to reviews if any topic-related keywords or form of topic-related keywords (e.g., keyword “learn” would have forms like “learning”) were present. As a starting point, we used keywords identified in the previous study (Early & Li, 2020). Note that multiple topics could be present within one review. We then fed reviews for each topic into the generation AI model ChatGPT, which we tasked with generating more keywords present in the reviews and related to the topic. We took the newly generated keywords and added them to the list of a priori keywords and assigned topics to reviews again. We did this three times, at which point all keywords the AI generated were already in use. After each addition of new keywords, the research team examined approximately 20 reviews for each topic to assess for false positives (e.g., the keyword “bonus” is identified as a term related to the topic “cost”, but how it is used in the sentence was talking about “bonus points for being a great center”). If a keyword added more false positives than true positives in the 20 reviews examined, we removed it from the list.[1]

The second iteration of assigning topics was only for reviews that still did not have any topics assigned to them. We fed these reviews into ChatGPT with the assignment to classify each topic as present or not present in each review. The AI was unable to categorize 4 percent of reviews (n=871; e.g., “Best daycare in town!”, “Very disappointed!”, and “Wow”). While some of these reviews truly do not relate to any of the topics we defined, others were too short for the AI to extract anything meaningful, so we were unable to categorize them. We omitted these reviews from analyses.

While topic modeling can tell us some of the content present in reviews, it cannot accurately tell us whether the reviews are positive or negative. For example, two reviews could be given the topic “warmth” with the keyword “friendly,” but with completely different opinions of the child care center (e.g., “staff are so friendly” and “staff are not at all friendly”). Therefore, we assigned a rating category of “positive” or “negative” to each review based on the reviewer’s star rating. Emulating Early & Li (2020), we defined positive reviews as those that rated the program as 4 or 5 stars and negative reviews as those where the program was rated with 1, 2, or 3 stars. For example, if the review included keywords that corresponded to the topic of warmth, we used the star ratings to determine whether that was a positive review of warmth (rated as 4 or 5 stars), or whether it was a negative review talking about warmth (rated as 1, 2, or 3 stars).

---

Footnote

[1] One keyword and five forms of keywords were removed from the search terms because they had too many false positives. The keyword “hand” was removed from the hygiene topic. From the academics topic, we removed the forms “teach,” “teacher,” and “teachers” stemming from the keyword of “teaching.” From the convenience topic, we removed the forms “time” and “times” stemming from the keyword of “timing.”