A Place for Mom is North America’s largest senior living referral service. Private-pay assisted living communities made up over half of our move-ins between 2012 and 2015. There is much debate among researchers, policymakers, and families about the benefits of private-pay assisted living to seniors and their families. We designed a Family Quality of Life Survey (FQLS) to measure how moving to an assisted living community affects the quality of life of the senior who moves and of their family member who helps them move. Here we describe the methods of the survey and its key results.
We found moving to an assisted living community is associated with greater overall quality of life for both the senior and family member. The move is also associated with greater quality of life across a range of dimensions including but not limited to: the senior’s nutrition and social well-being; the family member’s level of stress about the senior; and the quality of the relationship between the senior and their family member.
Because we did not have the resources to do a longitudinal analysis, we rely on two less direct survey methods:
We asked both moved-in and actively-searching families to rate the following aspects of their quality of life over the last three months on a five-point scale from (in most cases) “Very Good” to “Very Bad.” Alternative scales are noted where applicable.
We also asked moved-in families to rate how those aspects of their quality of life had changed compared to the year before the senior moved into assisted living. Again we used a five-point scale from “Much Better” to “Much Worse” (or “Much Healthier” to “Much Less Healthy” for diet and exercise level.
Again for both types of family, we asked about the following aspects of their relationship with the senior over the last three months. Alternative scales are noted where applicable; otherwise the five-point scale goes again from “Very Good” to “Very Bad.”
We also asked moved-in families how these factors changed compared to the year before the senior moved into assisted living, again using the five-point “Much Better” to “Much Worse” scale.
We also asked both types of family about the following components of the senior’s quality of life over the last three months. Family members rated each on a five-point scale from “Very Good” to “Very Bad.”
Again, we asked moved-in families to rate how these quality of life components had changed compared to the year before the senior moved into assisted living.
We asked families the following questions about their perception of senior living communities before and after moving in:
We asked the following demographic questions of respondents, which were taken from the list of SurveyMonkey certified questions:
We asked respondents if they had any trouble understanding the survey (possible answers were none, some, or a lot). We also asked respondents if we could contact them in a follow-up survey on this topic.
We only had resources for an online survey. Because many seniors would not be inclined to respond to an online survey, we built an instrument for family members who called A Place for Mom on behalf of a senior family member. We did not ask seniors about their own quality of life; instead we asked their family members to answer questions about the senior.
The survey instrument was internally validated through conversations with APFM VP of Partner Services Dan Willis, who has many years of experience in the senior living industry. Mark Scott from Sage Projections consulted with us on question design. A Place for Mom blog editor Gerard Gravallese edited the survey for content and clarity.
Survey scripts are available in PDF format upon email request to firstname.lastname@example.org.
Following are the constraints placed on the sampling frame as encoded in the query to the A Place for Mom transaction database for moved-in and actively-searching families.
This query served as the initial sampling frame, which we then used to build a sample that matches moved-in families to similar actively-searching families. Because we are a privately held company, we cannot disclose the size of the initial sampling frame.
To make more confident causal claims based on case/control data about how moving to assisted living influences quality of life, we used nearest-neighbor propensity-score matching. This method attempted to match each member of a moved-in sample to a similar member of the initial actively-searching sampling frame. The goal was to maximize the balance of the sample, i.e., make the distribution of propensity scores as similar as possible between cases and controls.
We successfully balanced the propensity scores of the matched sample based on a visual inspection of the histogram of post-matching propensity scores comparing cases to controls. The result is a matched sample of 4,133 moved-in families matched to an equal number of actively-searching families for a total of 8,266 potential respondents. Below we describe the steps in the matching procedure.
The target propensity scores estimate the probability that a family is a moved-in family given a set of predictors. The predictors we used are:
For predictors that were allowed to be missing when assembling the initial sampling frame, a new category “Unknown” was added.
We chose these predictors because: (a) they are nearly all of the predictors available in our database, and (b) they are predictors we know are likely to be important to predicting move-ins based on prior analysis at A Place for Mom (i.e., machine-learning models that predict the probability of a move and the time elapsed from contact to move-in), and that are often available for actively-searching leads even if they have not yet been referred to a senior living community.
The model must balance the need for capturing possibly complex interactions among predictors against the desire to avoid over-fitting, and should be optimized with respect to a sample-balance criterion. We used the twang package for the R statistical programming language to optimize a gradient-boosted trees model (as implemented in the gbm package, which twang extends). The model predicts the conditional probability of being a moved-in family given the predictors outlined in Step 1. We optimized the model’s tuning parameter (number of trees) with respect to the mean of the absolute standardized bias across the predictors. We tested a grid of equidistant values between 1 and 10,000 for number of threes. Because we are interested in the effect of moving in on families who actually moved in, the absolute standardized bias was calculating based on the average treatment effect among the treated.
After tuning the propensity-score model in Step 2, we use the tuned model to predict the propensity scores of every potential respondent in the initial sampling frame, regardless of their move-in status. That is, each potential respondent to the survey is assigned a probability based on the model that they are a moved-in family.
To assess balance, we looked at the histograms comparing the propensity scores of moved-in and actively-searching families both before and after matching. If the histograms after matching look similar across move-in status, the matching procedure is successful. The more similar the histograms, the better.
We selected at random about two-thirds of the matched sample, about equally split between actively-searching and moved-in families, to contact for the Family Quality of Life Survey. The rest were reserved for other survey activities this year. The sampling was split into three waves: a small pilot wave (sample size of 497) to get an early read on response rate, followed by two larger waves (sample size of 2,504 and 2,503 respectively). Within each survey wave and for each family type (moved-in versus actively-searching), there were three separate email survey collectors that used one of the following email subject lines:
The pilot wave spread the sample equally across the three alternative subject lines. We then used methods developed at Google by Steven L. Scott to estimate the probability that each subject line had the greatest response rate. In the next wave, we allocated the sample in proportion to these probabilities, and did the same for the final wave of the survey. A full analysis of response rates can be done upon request. Here we report the overall response rate (using only 100% completed surveys), which is about 6% after rounding to the nearest whole percent. The final completed-survey sample size is 294 completed surveys, 73% of which are moved-in families.
Survey waves were run for three days to a week each before sending a reminder, and then run for an additional 2-4 days before starting up the collectors for the next wave. All collectors remained open from the first collector’s open date (3 August 2016) to the close of the survey (25 August 2016).
Because the extreme high and low responses to quality-of-life and previous-perception variables tended to have low cell counts, we collapse the top and bottom two categories for analysis. An added benefit to collapsing the outcome variables is that the results are easier to interpret.
One of our questions about the senior’s perception of senior living communities matches a question asked in a Pew Research survey of older adults. To demonstrate that the seniors in our sample are similar to those in a scientific poll from a well-respected polling firm, we estimate the posterior distribution of the share of respondents answering in a particular way using independent Dirichlet-multinomial models with uniform prior distributions (i.e., prior concentration parameters all equal to one), one for each of the two surveys. Then we plot the posterior means and 95% credible intervals of the predicted shares.
In addition to measuring seniors’ perceptions of senior living using the Pew survey question, we use independent beta-binomial models, again with uniform distributions, to estimate the share of respondents who claimed each of the five possible causes for delaying their search for senior living.
Finally, we use a Dirichlet-multinomial model with uniform prior to measure the share of respondents with positive, lukewarm, negative, or no opinions in response to the question about their previous perception about senior living before calling A Place for Mom.
Ultimately, we need to look at how quality-of-life metrics vary together as well as with predictors from the survey and the APFM database, but our first analysis builds simple models to:
These estimates are based on independent Dirichlet-multinomial models with flat uniform priors, from which we derive 95% credible intervals for the share of respondents in each category (and, where applicable, family type) for each question, and where applicable by move-in status. For this exercise, we use the collapsed versions of the quality-of-life outcome variables, but only after simulating the posterior distributions of the share of respondents in the original categories.
Using collapsed outcome categories as described in a previous section, we use the flexmix package in R to build polytomous latent class regression models to see how factors like family-member age, finances, care needs, and ethnicity influence change in quality of life after moving to assisted living. We also add current quality of life as a potential predictor. The objective of the analysis is two-fold:
Only variables with sufficient variation within move-in status are chosen as potential predictors. When fitting models, we consider all possible models with main effects and (where applicable) first-order interactions with move-in status, insofar as those models are computationally tractable (i.e., the EM algorithm that the packages uses converges, the algorithm runs without error, etc.). We also fit intercept-only models. We use flexmix‘s built-in multinomial regression specification (implemented with function FLXMRmultinom) as the driver of the latent class regression.
The number of groups and the final predictors are chosen using two criteria:
We choose only one final model but retain the BIC values of each competing model.
The optimal propensity-score model was based on 4,846 tree-building iterations. The five predictors with highest relative importance averaged across trees were (in descending order): whether resident living in rehab/nursing facility, resident’s current living situation, desired state, contact’s state of residence, and the family’s stated senior-living budget.
Figure 1 compares the propensity-score distributions of moved-in and actively-searching families before and after nearest-neighbor matching. Note the considerable increase in balance evident in the greater similarity in histogram shape after matching. The matching process yielded 4,132 moved-in families matched to 4,124 actively-searching families for a total of 8,256 families. The section on survey waves and response rates describes how roughly two-thirds of this sample was split into three separate waves.
Figure 2 shows that the FQLS sample is very similar to the Pew Research sample in terms of the ordering of preferences and their relative weights. The gray lines represent the 95% credible intervals. Both surveys show that about three out of five seniors would rather stay at home with someone to care for them if they could not live on their own. Only about one in five families would move into an assisted living facility. One difference between the FQLS and Pew samples is that the latter estimates a larger difference in share of respondents who would move into an assisted living facility versus with a family member. Another is that the Pew Survey estimates a much larger proportion who who rather move into a nursing home, although for both surveys that share is quite small.
Figure 3 shows the proportion of families who checked each of the five possible reasons they delayed their senior living search. Three out of four families delay their search because the senior didn’t want to move. Eighteen percent of families delay their search because they didn’t want their senior loved one to live in of “those places.” Eighty-five percent of families checked one or both of those options.
Figure 4 shows that as many families have lukewarm feelings about senior living before calling as have positive feelings. Lukewarm and negative sentiment together represent half of families.
For each quality-of-life measure, we plot the comparison between moved-in and actively-searching families next to the distribution of moved-in families’ responses to questions about change in quality of life. Then we interpret the results with reference to additional summary statistics. The bars in the plots represent the posterior mean share of respondents. The light gray lines represent the 95% credible intervals.
The results below suggest that overall quality of life improves greatly for seniors when they move to assisted living. The effects are greatest and most certain for the quality of the senior’s nutrition and their social well-being. Less so for social well-being and physical health.
Seniors who moved are 70% more likely to report a good or very good overall quality of life than seniors who are still searching. They’re also 65% less likely to report a bad or very bad overall quality of life. Seventy-three percent of families who have moved report improvement in the senior’s overall quality of life, which is five times as many families who see overall quality of life worsen and six times as many families who see no change. Not only are seniors much more likely to see overall quality of life improve than worsen; they are also much more likely to see it improve than stay the same.
Seventy-eight percent of families who moved in report good or very good nutrition for the senior and are over 1.5 times as likely to do so than actively-searching families. They’re also nearly 40% less likely to report bad or very bad nutrition compared with actively-searching families. Seventy-three percent of moved-in families report improvement in the senior’s nutrition, which is eight times as many as families who say nutrition worsened and four times as many as who say nutrition remained the same.
Forty-eight percent of moved-in families report good or very good social well-being for the senior, which is over twice as many as the actively-searching families who do so. Moved-in families are also nearly 70% less likely to report bad or very bad social well-being compared to actively-searching families. Meanwhile, 64 percent of moved-in families report improvement in the senior’s social well-being, which is over four times as many as who say social well-being worsened and about three times as many as who say it stayed the same.
The differences between moved-in and actively-searching families are not as clear when it comes to the senior’s emotional well-being. Yet moved-in families are over twice as likely to see emotional well-being improve than worsen, and nearly 1.5 times as likely to see it improve than stay the same.
The effect of moving to senior living on the senior’s physical health is neither as large nor as statistically certain as for other measures, such as overall quality of life and nutrition. In fact, it’s highly uncertain whether seniors are more likely to see physical health improve than stay the same. However, roughly twice as many respondents report better physical health for the senior than worse.
The benefit of moving to assisted living on family members’ overall quality of life is less certain than it is for the senior. Yet there is moderate evidence that family members of moved-in seniors feel their quality of life improve. Moreover, there is strong evidence that the move alleviates the stress that family members feel about the senior. In addition, many families feel their social well-being improve, their diet and exercise level become more healthy, or caregiving have less of an impact on their ability to work. Financial well-being is most likely to stay the same after moving.
The comparison of overall quality of life between moved-in and actively-searching family members is uncertain. Yet 60 percent of family members of seniors who moved feel their overall quality of life improved, which is four times as many as who felt it worsen and twice as many as who felt it stayed the same.
Actively-searching families are 1.6 times more likely to feel high or very high levels of stress about the senior when compared to moved-in families. Moved-in families, on the other hand, are over five times more likely to feel low or very low levels of stress about the senior compared to actively-searching families. Sixty-four percent of moved-in families feel less stress about the senior than before the move, which is three times as many as those who feel stress increase. Family members are also 1.5 times as likely to see stress about the senior improve as increase or stay the same.
The difference in social well-being between moved-in and actively-searching families is unclear. Moreover, the majority (63 percent) of family members of seniors who move in feel no change to their social well-being. Yet nearly a quarter of family members feel their social well-being improve, nearly twice as many as those who feel it worsen.
There is no clear distinction between moved-in and actively-searching families in the family member’s diet and exercise level. The majority (60 percent) of families of seniors who move report no change in their diet or exercise levels. Yet nearly a third of moved-in families see improvement, which is over three times as many as those who feel their diet and exercise level get less healthy after the move.
There is no measurable difference between moved-in and actively-searching families on self-reported impact of caregiving on the family member’s ability to work. Yet 39% of moved-in families see improvement on this measure compared to only 8% who feel greater impact of caregiving on work after the move.
There is no measurable difference in financial well-being between moved-in and actively-searching families. Moreover, 70% of moved-in families feel no change to their financial well-being, and roughly equal shares of families feel it worsen (14%) or get better (15%).
There is some evidence that moved-in families may be more likely to have a good or very good relationship with the senior because there are fewer who report that the quality of the relationship is fair. Yet the magnitude of this effect is uncertain. On the other hand, half of families report a better relationship with the senior, which is twice as many as those who see no change and five times as many as those who see the relationship worsen.
We tested all possible latent class models that included demographic variables, previous perception of assisted living, number of types of assistance with daily activities needed, and stated family budget. For the analysis of change in quality of life, we also included current quality of life as a predictor. For the analysis of current quality of life, we also tested all possible models containing at last move-in status and at least one interaction of move-in status with another variable. For each regression specification, we tested models with between one and five latent classes.
Intercept-only models with two latent classes had the optimal BIC scores for both current quality of life and change in quality of life. We do not report the current-quality-of-life model here. Instead, we focus on how moved-in families can be grouped based on their responses to questions about changes in their quality of life.
Broadly speaking, the two latent classes are characterized by the propensity of their members to report improvement in quality of life for both the senior and the family member. Consistent with the simpler analyses in the previous section, positive outcomes for the senior are more common than for the family member even in the positive-outcome group.
Figure 29 shows the share of respondents in a given group who give a particular rating on each question about the senior’s quality of life. Note how nearly all of the members of the positive-outcome group report improvement in the senior’s overall quality of life. Consistent with the simpler analyses in the previous section, nutrition and social well-being are also more likely to improve for the positive-outcome group. Interestingly, half of the members of the negative-outcome group report improvement in the senior’s nutrition.
Figure 30 shows the breakdown by latent class for questions about the senior’s family member. More than three quarters of the positive-outcome group report improvement in the family member’s overall quality of life. An even larger number of these families report improvement in the level of stress they feel about the senior, and three in four family members report improvement in the quality of their relationship with the senior.
Based on these results, we can come to the following conclusions about how a move to assisted living influences quality of life.
It makes sense that a move to assisted living is more likely to improve the senior’s quality of life than their family member’s. After all, the primary goal of an assisted living community is to help seniors thrive despite their need for assistance with activities of daily living and advancing age. Benefits to the senior’s family are secondary.
The task now is to repeat this survey in subsequent years to increase sample size. An increased sample size would allow us to assess how the impact of moving to assisted living varies by caregiver demographics, senior’s care needs, financial situation, living situation, previous perception of senior living, and other variables. Doing so will help consumers, policymakers, and the industry understand which consumer segments benefit most and least from a move to assisted living.