This is an open-access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in the Journal of Medical Internet Research, is properly cited. The complete bibliographic information, a link to the original publication on https://www.jmir.org/, as well as this copyright and license information must be included.
Web-based reviews of physicians have become exceedingly popular among health care consumers since the early 2010s. A factor that can potentially influence these reviews is the gender of the physician, because the physician’s gender has been found to influence patient-physician communication. Our study is among the first to conduct a rigorous longitudinal analysis to study the effects of the gender of physicians on their reviews, after accounting for several important clinical factors, including patient risk, physician specialty, and temporal factors, using time fixed effects. In addition, this study is among the first to study the possible gender bias in web-based reviews using statewide data from Alabama, a predominantly rural state with high Medicaid and Medicare use.
This study conducts a longitudinal empirical investigation of the relationship between physician gender and their web-based reviews using data across the state of Alabama, after accounting for patient risk and temporal effects.
We created a unique data set by combining data from web-based physician reviews from the popular physician review website, RateMDs, and clinical data from the Center for Medicare and Medicaid Services for the state of Alabama. We used longitudinal econometric specifications to conduct an econometric analysis, while controlling for several important clinical and review characteristics across four rating dimensions (helpfulness, knowledge, staff, and punctuality). The overall rating and these four rating dimensions from RateMDs were used as the dependent variables, and physician gender was the key explanatory variable in our panel regression models.
The panel used to conduct the main econometric analysis included 1093 physicians. After controlling for several clinical and review factors, the physician random effects specifications showed that male physicians receive better web-based ratings than female physicians. Coefficients and corresponding SEs and
This study found that female physicians receive lower web-based ratings than male physicians even after accounting for several clinical characteristics associated with the physicians and temporal effects. Although the magnitude of the coefficients of
Web-based reviews of physicians have been gaining significant popularity among health care consumers or patients over the past 2 decades. Some examples of popular websites for web-based physician reviews are RateMDs [
The literature on web-based reviews of physicians has been growing in the past 10 years. Using data from the United States and other countries, numerous studies have examined the content and valence of web-based physician or hospital reviews and the factors that could explain their variance [
Another substream has investigated the influence of web-based physician reviews on patients’ choices. There has been a significant interest among health care researchers and practitioners in the health care consumers’ awareness of web-based physician reviews [
The increasing reliance on web-based physician reviews is indicated by other surveys also [
As web-based health care information, including physician reviews, is publicly available and easily accessible, there has been a long-standing concern among the health care providers and research communities about the quality and clinical relevance of web-based health care information [
There has been a long-established interest among researchers in the impact of physician gender on patient communication and patients’ choice of physicians. Extant literature has found that female physicians tend to engage in patient-centered communication [
Questions about whether patients have a preference for male physicians over female physicians, and vice versa, and whether their opinions of physicians are affected by the physicians’ gender have also received substantial attention from health care professionals and researchers. For instance, in a survey of 185 patients, Fennema et al [
With the proliferation of web-based physician reviews among patients or health care consumers, a natural and important question is, “Whether and to what extent is a physician’s gender related to their online reviews after accounting for patient risk and time shocks (time fixed effects)?”
After a careful review of the existing literature, we found that the potential effect of physician gender on web-based reviews of physicians has not received sufficient attention. In the few studies that have examined the relationship between physicians’ gender and their web-based reviews, the findings have been mixed. For example, Dunivin et al [
In the examination of the aforementioned relationship, it is important to account for the characteristics of patients, such as patient risk, in some form. It is also important to account for the variation in the reviews over time to determine the direct relationship between physicians’ gender and their web-based reviews. Including patient risk allows us to account for the health characteristics of a significant patient population under the care of physicians. Not controlling for such characteristics can potentially bias the results because a physician’s interaction can be affected by the existing health condition of their patients. Therefore, we examine the effect of physician gender on web-based patient reviews, while controlling for patient health risks over time.
To the best of our knowledge, our study is among the first to examine the effect of physicians’ gender on their web-based reviews over time and after accounting for patient risk. Furthermore, our study is the first to conduct such an investigation using physician data across Alabama, a state that has received very little attention in the literature on web-based physician reviews. We accomplish our analysis by using a unique data set that we created by combining data from web-based physician reviews from a popular physician review website, RateMDs, and clinical data from the Center for Medicare and Medicaid Services (CMS) for the state of Alabama.
No ethics board review or approval was required for this study. All the raw data that were collected for this study are publicly available on the web.
To study whether web-based reviews of physicians are more favorable toward male or female physicians, we constructed a panel data set of physicians in Alabama using data from 2 sources. The unit of analysis in our study was a physician, and the time periods in the panel were years. We collected data on web-based reviews and the gender of physicians from RateMDs to construct our web-based review data set spanning from 2012 to 2018. We used Python (Python Software Foundation) to collect data from RateMDs. We also obtained clinical data on physicians from Medicare Provider Utilization and Payment Data: Physician and Other Supplier [
Each physician in our final panel has a unique national provider identification number that was collected from CMS. This ensured that all the physicians in our final panel were unique.
Example screenshot of RateMDs reviews for a physician.
As we were examining whether the web-based reviews of physicians are favorable to male or female physicians, we constructed our dependent variables using the numeric physician ratings from RateMDs. Physicians on RateMDs can be rated on four dimensions: helpfulness, knowledge, staff, and punctuality. The ratings for each of these dimensions are on a scale of 1 to 5, with 5 being the best possible score and 1 being the lowest score. To capture the information in each of these four dimensions, we constructed the following four dependent variables:
Our key explanatory variable was a time-invariant variable,
To construct these topics (latent topics), we used topic modeling techniques based on Latent Dirichlet Allocation (LDA) [
We created a corpus of all the reviews using an R text-mining package(TM) within RStudio, after which we converted the corpus to lower case [
care, doctor, staff, recommend, patient, time, knowledg, help, friend, love, wonder, high, listen, excel, and feel
time, office, doctor, wait, staff, patient, appoint, call, nurs, rude, visit, day, question, hour, and talk
doctor, surgeri, pain, care, medic, life, patient, treat, recommend, time, day, surgeon, procedur, treatment, and feel
We had physicians from across 34 specialties in our final panel data set. The 15 specialties with most physicians (in descending order of the number of physicians) were as follows: general (family) practice, obstetrics and gynecology, internal medicine, orthopedic surgery, neurology, otolaryngology, cardiology, ophthalmology and optometry, psychiatry, dermatology, general surgery, podiatry, urology, endocrinology, and rheumatology. Physicians in these 15 specialties accounted for approximately 85.73% (937/1093) of all the physicians in our panel data set. Table S1 in
We used physician random effects panel regression, along with year fixed effects to account for time shocks. A time shock in the context of this paper can be considered as an event or collection of events that can impact physicians across the board in the duration of a year. For example, a statewide or nationwide health care policy change would likely have an impact on physicians across different specialties. As the analysis used panel data, it was important to account for such time shocks. We did so by including year fixed effects in our regression specifications. We used Stata (StataCorp) for conducting our econometric analysis.
We leveraged the physician random effects model instead of the physician fixed effects model to estimate the effect of physician gender because of the following reasons: (1) our main explanatory variable,
Distribution of total number of physician reviews across years.
Comparison of average overall ratings for female and male physicians across years.
Comparison of average helpfulness ratings for female and male physicians across years.
Comparison of average knowledge ratings for female and male physicians across years.
Comparison of average staff ratings for female and male physicians across years.
Comparison of average punctuality ratings for female and male physicians across years.
Descriptive statistics (number of observations=3446).
Variable | Values, mean (SD) | Values, median | Values, minimum | Values, maximum |
|
3.64 (1.43) | 4.25 | 1 | 5 |
|
3.54 (1.65) | 4.37 | 1 | 5 |
|
3.74 (1.54) | 5 | 1 | 5 |
|
3.69 (1.48) | 4 | 1 | 5 |
|
3.60 (1.49) | 4 | 1 | 5 |
|
0.41 (0.45) | 0 | 0 | 1 |
|
0.27 (0.40) | 0 | 0 | 1 |
|
0.32 (0.42) | 0 | 0 | 1 |
|
1.23 (0.41) | 1.14 | 0.53 | 5.62 |
The coefficient of
Estimation for OverallRating (N=1093)a.
Variable | Coefficient (SE) | |
|
−0.162 (0.060) | .007 |
|
−0.056 (0.086) | .52 |
|
1.557 (0.058) | <.001 |
|
0.739 (0.071) | <.001 |
aSpecialty controls=yes; year fixed effects=yes; robust SE=yes; overall R-squared=0.267; within R-squared=0.168; between R-squared =0.339.
Estimation for HelpfulnessRating and KnowledgeRating (N=1093).
Variable |
|
|
|||
|
Coefficient (SE) | Coefficient (SE) | |||
|
−0.185 (0.069) | .008 | −0.198 (0.065) | .002 | |
|
0.003 (0.098) | .97 | −0.057 (0.094) | .54 | |
|
1.702 (0.069) | <.001 | 1.492 (0.064) | <.001 | |
|
0.688 (0.084) | <.001 | 0.513 (0.080) | <.001 |
aSpecialty controls=yes; year fixed effects=yes; robust SE=yes; overall R-squared=0.239; within R-squared=0.153; between R-squared=0.310.
bSpecialty controls=yes; year fixed effects=yes; robust SE=yes; overall R-squared=0.220; within R-squared=0.137; between R-squared=0.282.
Random effects panel regression (StaffRating and PunctualityRating; N=1093).
Variable |
|
|
|||
|
Coefficient (SE) | Coefficient (SE) | |||
|
−0.095 (0.062) | .13 | −0.172 (0.067) | .01 | |
|
−0.045 (0.087) | .61 | −0.127 (0.105) | .23 | |
|
1.547 (0.063) | <.001 | 1.488 (0.063) | <.001 | |
|
0.923 (0.076) | <.001 | 0.832 (0.074) | <.001 |
aSpecialty controls=yes; year fixed effects=yes; robust SE=yes; overall R-squared=0.247; within R-squared=0.155; between R-squared=0.315.
bSpecialty controls=yes; year fixed effects=yes; robust SE=yes; overall R-squared=0.234; within R-squared=0.130; between R-squared=0.318.
We added additional control variables to check whether our findings would change. The three additional variables were
We conducted further robustness checks by removing the specialties in our panel in which both genders were not represented. This helped us mitigate the concern that a possible bias may arise owing to the absence of physicians of one of the genders in any of the specialties in our panel. The results displayed in Tables S5-S7 in
In our next robustness check, we conducted our main regression analysis without topic controls. This test was conducted to examine whether the topic variables may have introduced a systemic bias in the specifications owing to the manner in which they were constructed and whether the negative coefficient of
In summary, we conducted three additional robustness checks as explained above: (1) included additional control variables, (2) removed the specialties that did not include physicians of both genders, and (3) removed the topic controls. After conducting these robustness checks, we can conclude that female physicians tend to receive worse web-based reviews than their male counterparts. This finding is consistent across the regression specifications used in this study.
A concern could be about how representative the data in our panel are of the original data collected from RateMDs and Medicare (CMS). To address this concern, we calculated the descriptive statistics of the variables shown in
Our study provides an important contribution to the growing literature on web-based physician reviews and physician gender. A possible concern could be that the differences observed in the reviews between physicians of different genders could be driven by the differences in the quality of care or outcomes delivered by physicians of different genders. To address this concern, we performed a substantial search of the existing literature examining the differences between the quality of clinical care or outcomes delivered by male and female physicians. We found several research papers in this context [
We found that male physicians receive better web-based reviews than female physicians after controlling for their clinical characteristics such as specialty and patient risk. Although the difference between the web-based ratings for male and female physicians was statistically significant, the average magnitude of the difference was not substantial. Our findings support that of Dunivin et al [
Our findings have important implications for health care researchers, professionals, and policy makers. First, the empirical evidence of web-based reviews is less favorable toward female physicians, after accounting or controlling for several clinical aspects (including specialty and Medicare patient risk), and temporal effects should inform health care professionals and policy makers that patients’ opinions are consistently more favorable toward male physicians than toward female physicians. This cannot be overlooked even though the magnitude of the effect of gender on web-based reviews is not sizable.
Gender bias in reviews has been reported across multiple domains, including academia. Murray et al [
Large societal-level aspects may also be in effect; however, that would seemingly be very hard to account for within a single portal. Sprague and Massoni [
Concentrated efforts to educate and inform patients about female physicians’ competence are needed. This can help to reduce implicit bias among patients toward the competence of female physicians compared with their male counterparts. These websites serve as an important resource for both reviewers and readers of the reviews, and the information needs to flow well. At the same time, readers of the reviews may be served better if the reviewers are asked to provide opinions about physicians of different genders before they provide a review for a physician. To solicit reviewers’ predisposed opinions about physicians of different genders, the questions can be framed in a manner that does not make the reviewers feel that they are being investigated for their opinions. After collecting their opinions on this issue, the websites may consider filtering the reviews provided by reviewers with an overt bias against physicians of one gender. The question of how to design the website to reduce the possible gender bias is complex and requires serious thought and consideration from both researchers and website designers. By leveraging previous research efforts targeted at informing users of bias potential, review portals can better collect and present information about physicians.
Our study has a few limitations. First, we constructed our patient risk scores using the HCC risk score from Medicare data. Although Medicare is among the largest health care payers or insurers in the United States, further studies can attempt to validate the findings of our study using clinical data from other insurers. For instance, a significant proportion of the patient population in the United States has insurance from private insurers. Future studies can attempt to validate our findings by constructing clinical variables, such as risk scores, using clinical data from one or more private insurers. Second, we focused on the physician data from Alabama. Although it is 1 state, it provides a good mix of rural and urban counties. Future studies could extend this work to other states and compare the findings across a broader set of patients and health care providers.
The findings of this study suggest that gender bias in web-based reviews needs to be examined more closely. Additional studies that identify factors impacting this gender bias could help us develop strategies to mitigate gender bias in web-based reviews. Given the shortage of health care providers and the need for a robust and diverse health care workforce, such studies can help not only the service providers but also policy makers, educators, and administrators. If the administrators of hospitals and clinics are made aware of this bias and acknowledge it accordingly, institutional changes can be implemented to support and empower women to take up more leadership roles in clinical settings. As Sandberg [
These focused efforts can provide a strong signal to patients about the competence of female physicians and, in turn, increase their confidence in the care provided by female physicians. This can further help to improve the overall care delivered to patients, as the increase in patients’ confidence can improve their communication with physicians, irrespective of the physicians’ gender. However, an open research question is whether the bias observed in web-based physician reviews is also observable in offline physician surveys. To examine this question, studies that compare reviews of male and female physicians in web-based and offline media need to be conducted.
Tables depicting the results of additional analysis including robustness checks.
Center for Medicare and Medicaid Services
hierarchical condition category
Latent Dirichlet Allocation
None declared.